The story below is AI fan fiction, but the companies that understand the patterns discussed will win the next decade.

In the spring of 2025, greenhouse producer Vermeulen Serres lost a customer of fifteen years.

In retrospect, the churn was avoidable. In 2024, when the wholesale tomato market took a downturn, Kees-Jan, owner of Groentenenfruit.nl, switched one of his large greenhouses to strawberries. The crop change went into Dynamics in fifteen seconds. What didn’t go anywhere was the implication. His heating system had been specced for tomatoes back in 2010, the crop change invalidated those calculations, and the only person who would have spotted it was Luc, the CTO of Vermeulen Serres. Six months after the switch, Kees-Jan logged a heating modification request. A junior employee answered that the existing system was within spec, and closed the ticket. The following February, under-temperature alarms set off during winter, two thirds of Kees-Jan’s strawberry crop failed, and he moved his account to a competitor.

Vermeulen Serres is a Flemish family business. Around 75 employees, 20 million in turnover, building greenhouse complexes for industrial growers. The ERP runs on Postgres. The CRM is a Microsoft Dynamics install where data goes to decay. Accounting was forklifted to a Peppol-compliant tool when the ERP couldn’t be updated in time for the e-invoicing requirement from the EU. Sales operations runs a parallel Excel pipeline “just for our team.” If you look under the hood of most mid-market companies in Europe and North America, you’ll find an IT mess. Yet almost nothing in the public AI discourse is written for them.

Vermeulen’s CEO was sure Luc would have caught Kees-Jan’s issue had he seen the project. Thirty years at the company, Luc’s title says technology, but his actual specialty is greenhouse engineering. Luc is the type of colleague you can randomly approach in the coffee corner about why the glassware spec was changed on the 2018 Verbruggen project and he will give you a precise answer from memory.

IT became part of his mandate because nobody else picked it up. The Kees-Jan incident was the first time the CEO identified the risk, but there would be others. Luc couldn’t be on every project. And Luc is headed for retirement in eighteen months.

The two clocks

Every system in your organization has two clocks. The state clock captures what’s true right now. It tells you who the customer is, whether a contract is active, what’s been delivered in a project, what the maintenance plan says etc… The event clock, if you were to build one, records not only what happened, but also in what order, and why. It answers questions like: which negotiations preceded the contract, which exceptions were granted, which decisions hold up and which can be forgotten. Most companies have decades of state-clock infrastructure. I don’t know any with an event clock.

The “context graph debate” that has dominated tech writing since January is really a debate about how to build the event clock. The losers will be the ones who treated “context graphs” the way they tackled “Agile” in 2008: a container term that meant something at first, then meant less, then meant whatever the speaker wanted it to mean that day. Two terms get conflated and worth pinning down. A knowledge graph captures the state layer: entities and their relationships. A context graph adds the event layer: causes and sequences. I’ll use “context graph” to mean the combined thing.

The engine and the car

So, does AI unlock context graphs?

Well; not in and of itself because models are probabilistic. They’re sophisticated stochastic parrots that approximate the right answer most of the time. To run a business on top of one, you have to wire the AI magic into deterministic software that knows when to defer to the model and when to use (or write) plain software. Garry Tan calls these the latent and deterministic sides of the AI coin. Latent is where the model judges under uncertainty: “is this email sales-relevant?” Deterministic is where code does what code does: “subtract January revenue from February.” The art is using each in its lane.

“This is the loop that makes the whole architecture work: the latent space builds the deterministic tool, then the deterministic tool constrains the latent space.” - Garry Tan, CEO of YC

Tan’s frame is best understood with a metaphor: the AI model - the latent space - is the engine, and you can’t drive anywhere with just an engine. Latent and deterministic space together make the car. The strategic assumption you can safely hold for the next decade is that the model advantage will decay faster than the organizational advantage around it. Put more simply: the engine is becoming a commodity.

Frontier models are converging across providers; some are already available locally as open source. Whatever moat your organization wants is going to live somewhere in the car. If this sounds like technical plumbing, I hate to break the news to you, but the plumbing is where the operating leverage will live.

Building the event clock

Vermeulen Serres has spent fifteen years postponing decisions about its IT systems; Luc’s retirement and the Kees-Jan incident remove the last political reason to keep deferring. The question is whether the company can encode enough of what Luc knows before he leaves, and whether it can spot the next Kees-Jan before he has the conversation.

The back office

Luc brings in an AI transformation engineer for the eighteen-month window. Her first call is the data classification: the company treats its data as proprietary but not regulated, which makes public clouds an option and unlocks half the tool stack (regulated industries face different choices).

The first build gives an agent supervised access to the company stack: the ERP, Dynamics, the project archive, supplier specs, the financials, the support inbox. Everything is read-only by default; writes go through scoped service accounts with audit logs. The agent answers questions like “which Q2 projects are running over budget on materials, and what changed?” without anyone in finance having to open a spreadsheet.

Letting systems talk to each other sounds not only difficult, but also boring. At least until you realize AI legibility will be the single biggest business advantage of the next decade. Despite all these IT systems, most companies are not legible to machines. Most companies are barely legible to their own people. Bear with me and assume Vermeulen Serres achieves the legibility milestone - we’ll come back to the challenges later.

Now that AI agents can read their systems of state, Vermeulen’s CEO wants to build the event clock: a system that would have caught Kees-Jan’s crop switch the moment it landed in Dynamics in 2024, instead of letting it surface eighteen months later as a churned account.

The logic sounds straightforward. A crop switch on a fifteen-year-old greenhouse should trigger a re-evaluation of the heating spec. Five other customers made a similar switch in the same window; an event-clock-aware agent would have flagged all five for engineering review before the cold snap hit.

The architecture the engineer chooses is a vector-graph hybrid. The vector layer (Postgres + pgvector, since they already run Postgres for the ERP) handles fuzzy retrieval like “find anything that mentions delivery delays at Kees-Jan’s site.” The graph layer (Graphiti) holds the explicit relationships: this client switched crops in this month for this reason, this support ticket was opened against this delivery, this engineer wrote the heating spec on this project. Bi-temporal queries become possible. “What did the support team know about Kees-Jan’s heating on the day his ticket was closed?” Now try that with SQL.

The temporal aspect is new: every fact carries two timestamps, one for when it became true in the world and one for when the company learned about it. The team can replay the graph as it stood on the day Kees-Jan logged his heating modification request and audit what the agent should have known then. The event clock is a record of what was knowable when. This turns the graph from a database into an audit trail, and gives the company the substrate of a system that updates its self-model based on what it observes. Cybernetics, but with 2026 tools.

The meeting record

The vector-graph hybrid only works if the events get captured. The state systems (ERP, Dynamics, project archive) capture transactions and documents, not conversations. Most of what Luc knows lives in conversations, like the call with Kees-Jan where he explained why he switched crops.

The current generation of meeting tools (Granola, Fireflies, Teams’ built-in transcription) solved the recording problem but they can’t connect the text to the structured data or resolve the entities. Kees-Jan’s call transcript lives in Granola, but his customer record lives in Dynamics and his project history sits inside Sharepoint. None of them know about the others.

The architectural approach is to make the transcript a first-class citizen in the graph. Extract the entities (Kees-Jan, the heating problem, the crop-switch reasoning) and connect them to the nodes that already exist. Add the bi-temporal edge: when Kees-Jan stated this, and when the company logged that he stated it. Now the multi-hop query from earlier gets a fourth hop: heating issue → crop switch → market reasoning → Kees-Jan’s own words on the call. The agent doesn’t just spot the pattern across five clients on the same crop trajectory; it can quote the customer back to the account manager preparing the renewal.

It’s also the answer to the pension question. The engineer can’t extract thirty years of Luc’s judgement retroactively. She can pipe every meeting he attends through the same extraction pipeline for the eighteen months before he retires. A useful share of what he knows ends up landing in the graph that way.

What’s in the context graph stack?

The greenhouse case touched four of the six layers an agent system needs. Most arguments about context graphs are different arguments stacked on top of each other, with the participants disagreeing about different layers without noticing. Let’s unpack them.

The data layer.
Most companies already have data layers: relational databases for structured records, document stores for unstructured content, search indexes for full-text retrieval. AI adds two new shapes. Vector retrieval finds things by similarity (embed the query, find chunks that mean roughly the same thing). Graph retrieval finds things by explicit connection (walk the typed relationships between entities). Memgraph and Neo4j are the household names competing to be the go to graph store. Vectors find the haystack; graphs find the path through it. Most production systems end up running both.

The capture problem. A lot of organizational knowledge lives in people’s heads. The state clock (who is the CFO, which plan is Kees-Jan on) is mostly captured by your existing systems. The event clock (why we awarded that off-list discount, why the NPS survey and the support tickets tell a different story) is mostly missing. Glean tries to surface relationships by indexing everything across your internal tools. Palantir embeds engineers who do the structuring by hand, but they only operate on government-sized organisations. Microsoft’s GraphRAG extracts knowledge graphs from documents automatically. These are different bets on how much human discipline is required.

The wiring. Agents need to call tools, read databases, and hand off to other agents. MCP (Anthropic’s Model Context Protocol) is the wiring standard for tools. A2A (Google’s Agent-to-Agent) is the wiring standard for agent-to-agent calls. Both are now Linux Foundation projects. Low-code automation platforms like n8n sit one layer up, connecting hundreds of services without code and adding AI reasoning at specific steps. That’s where most companies without a platform team start. MCP usage itself is already shifting toward Anthropic’s Code Mode pattern, where tool definitions live as code the agent imports rather than as context it loads upfront, dropping token costs by an order of magnitude.

The cockpit. Someone has to give the agent instructions and read the result. Engineering teams gravitate to code-first environments like Cursor or Claude Code. Some knowledge workers are turning to the terminal to get the most out of their harness; others are limiting themselves to Claude Cowork. The jury is still out on what AI UX will look like and that question won’t get answered while the other layers are still moving.

Memory. AI chatbots have unreliable memories. An agent that handles your customer accounts has to remember all the details of Kees-Jan’s preferences. Mem0 extracts and reconciles facts as conversations happen. Graphiti (Zep) keeps the full history with four timestamps per relationship. Valid_at and invalid_at track the event time, when things happened in the real world. Created_at and expired_at track the system time, when the company learned about the events. Letta (formerly MemGPT) lets the agent manage its own memory the way an operating system manages RAM. These are different bets on what “remembering” actually means.

The harness. The harness is the software wrapped around the model - the car that makes the engine into something that can get you from A to B. Claude Code, Cursor, and OpenClaw are harnesses. LangGraph and CrewAI are orchestration frameworks that coordinate multiple agents. Memory and harness are not separable design decisions. The harness is what controls how memory enters and leaves the agent’s context, and which APIs that data passes through on the way.

Why this is hard

Building any of this is hard on different levels.

The social problem is well-known by now. The horticulture company has people who have been doing their jobs for twenty years. Some of them are excited about AI. A larger group believes that AI threatens the parts of their work they were proudest of, and they aren’t coming to workshops in good faith. A smaller group is already using their personal ChatGPT in the evenings and won’t tell anyone. The CEO has to introduce tooling that visibly redistributes work, and probably people, while the company is trying to grow. None of this is a context-graph problem, but it determines whether the context graph gets built.

The technical problem is the inheritance tax of the existing stack. The ERP runs on a Postgres version that hasn’t been upgraded in years. Dynamics is on the cloud version, but the parallel Excel pipeline in sales operations carries half the deals nobody else can see. The project archive is fifteen years of unstructured Word and PDF files, named inconsistently. There is no data governance function, deployments still route through the IT manager, and infrastructure as code is on next year’s roadmap. Building a production context graph against this substrate means doing the data-quality work that nobody has done for fifteen years, and doing it while the agent is already running in production.

Then there is a deeper, sociotechnical problem. Organizations don’t reliably know what they do, experts pattern-match in ways they can’t articulate, and most of the decisions that get filed in your CRM are post-hoc stories the company tells itself about events that happened. Capturing those stories at scale gives you a high-fidelity record of organizational fiction.

Underneath all three sits a bootstrapping problem. The companies that need a context graph most are the ones running on tribal knowledge, and tribal knowledge is exactly what can’t be queried. Discipline isn’t a byproduct of capture. It’s the other way round.

The prize is bigger this time

Despite the fog of war, one thing is certain: some companies will crack this faster than others. We saw it with agile software development, and again with DevOps. The early adopters built moats while the laggards held meetings. History will repeat.

The size of the prize is what’s different this time. Foundation Capital frames context graphs as a trillion-dollar opportunity, and the order of magnitude is probably right. Agile laggards got slower software cycles, DevOps laggards got worse uptime, and AI laggards risk being out-competed by companies with five times their operational leverage. Companies that build a working context graph end up in a different category from their peers. The historical analogy here is closer to electrification than to agile.

One last strategic point. Switching your agent’s model is straightforward; the APIs converge and the capabilities are comparable. Switching your agent’s memory layer is not. Proprietary formats coupled to vendor harnesses, accumulated organizational context that doesn’t export. Most other infrastructure is scaffolding over current model limits and will be absorbed; memory won’t be. Whoever controls your memory layer controls the lock-in.

In Part 3, I’ll argue that the discipline these systems require comes from AI adoption itself. Teaching an agent how your business works is the same act as making the business legible to itself for the first time. The context graph and the operating model end up being the same artifact, viewed from two angles.

You’re in a board room. It’s December 2026.

The board room belongs to a company called Cynalco Medics. They make a wearable patch called Meridian. It sits on your upper arm and reads glucose, cortisol, and heart rate variability continuously. The hardware is a marvel: clinical-grade sensors, three-week battery, medical adhesive that doesn’t irritate skin. It predicts metabolic crashes and stress events before they happen, then coaches you through them in real time.

The quarterly review is underway. Around the table: CEO, CFO, VP of Product, VP of Hardware, the clinical lead. In the corner of the conference display, a small waveform pulses gently. Everyone calls her Audrai. She’s the company’s agent layer, connected to the operating model, all systems of record, 12 months of meeting transcripts, chat history, the clinical trial database, even customer support tickets.

The CEO asks about Meridian 2.0. The launch slipped by five weeks.

Before anyone speaks, Audrai’s voice comes through the room’s speakers. “The delay wasn’t in the firmware or the companion app. Those were ready by October 15th. The timeline change can be traced to the insight layer. The original recommendation engine was flagging cortisol spikes too aggressively. Users got twelve alerts a day. In early feedback, sixty-three percent said they were ‘overwhelming’ or ‘not actionable.’”

No one asked Audrai to research this. She’d already pulled the context before the meeting started.

The VP of Product nods. “We rebuilt the personalization model. Took three weeks to get the false positive rate down.”

Audrai adds: “And another two weeks for the clinical team to validate the new thresholds. I have the exact handoff logs if you want them.”

In 2024, this scene would have gone differently. The product lead would mention “unexpected complexity.” The hardware VP would note that the sensors were ready on time. Someone would invoke “cross-departmental dependencies” as a catchall. Thirty minutes of corporate choreography, shuffling blame without assigning it.

But it’s the end of 2026, so the facts are on the table.

The CFO leans forward. “Audrai, what’s our current position on unit economics?”

“Better than projected. The extra five weeks let us ship with the refined model. Early cohort NPS is 71, versus 54 for the original version. Support tickets per user are down 40 percent. If those numbers hold, customer acquisition cost recovers in four months instead of seven.”

The room is quiet for a moment.

“So the delay was the right call,” the CEO says.

“The delay wasn’t a call,” Audrai replies. “It was a discovery. You didn’t know users needed a calmer AI until the first build proved the opposite. The time wasn’t lost. It was spent learning what the product actually needed to be.”

The meeting ends eighteen minutes early. There’s nothing left to debate.

That scene isn’t real. Not yet.

But it’s late January 2026, and I’m placing this prediction on record. Apologies for the timing. “Predictions for 2026” posts are supposed to land in December. I’ve been busy building the systems I’m about to describe. More on that later.

Marc Andreessen was right, but early. Software is eating the world, but we haven’t seen nothing yet. We are in the middle of the Youtube moment for software. The amateurs are coming and they are vibecoding patches for every software-shaped hole they can find. This also means a lot of smart coders will find new hard problems to work on, from hardware to ASICs.

The Cynalco boardroom is a projection, maybe twelve months out, maybe eighteen. The specific details will be wrong but I have unwavering conviction that the trajectory is correct. Every component already exists: voice interfaces that feel natural, agent layers that can query operating models, context architectures smart enough to hold months of company history.

Companies that treat this as a 2028 problem may not have a boardroom to sit in by 2027. The gap between “experimenting with AI” and “running on AI” is closing quickly, and the organizations on the wrong side of that gap will find themselves outmaneuvered by leaner, faster competitors who made the switch earlier.

So what has to change between now and the boardroom scene? A few things, in rough order.

First: the smallest value-adding unit shifts from the team to the individual.

The smallest unit

For decades, teams were the smallest value-adding unit of any organization. Specialized contributors created value through coordination. Methodologies mattered because they governed how teams moved together. Agile, Scrum, SAFe: operating systems for groups of humans trying to build software without stepping on each other’s toes.

These methodologies are obsolete, but the consultants don’t know it yet.

Individual contributors who have learned to direct swarms of agents will have a disproportionate effect on an organization’s success. One person with judgment and a well-tuned fleet of AI tools can outpace a ten-person team operating the old way. We’ll see a rise in solopreneurs and it’s going to be a tough time to be an incumbent.

For most of industrial history, individual contributor performance was normally distributed. Most people clustered around the middle. The difference between a good employee and a great one was meaningful but bounded. You could build organizations around averages because averages were predictive.

With AI, that distribution is becoming a power law (there is emerging science to back up this claim).

In a power law, a small number of individuals account for most of the output. The gap between the median and the top widens dramatically. A single person with the right capabilities can now do what used to require a team, a department, a small company. The “10x engineer” was already a cliche. We’re entering the era of the 100x employee.

This isn’t about replacing employees with AI (not yet, at least). I believe we are about to see a new category of employee: the hired gun who shows up with a personal software stack that compounds every task they touch. They fix friction points; automate workflows, and accumulate leverage the way investors accumulate returns. These people become irreplaceable. And the organizations that enable them will pull ahead of those that don’t.

The other side of this distribution is harsher. In a power law, long tails are long. People who refuse to adopt and treat AI as a threat rather than a tool, will find themselves competing for a shrinking pool of roles where the old rules still apply.

The ancestral lizard brain recoils from change by default. Most people will resist AI, and their resistance will cost them. Change management has always been the bottleneck in organizational transformation. Human rewiring is often harder than technical change. But the math is shifting: when your AI-native competitor can ship in a week what takes you a quarter, “we’ll get there eventually” stops being a viable strategy.

Teams without the burden of change management—whether because they’re new or because leadership forced the transformation—gain the most advantage during generational platform shifts. You’re all going to have to ADKAR faster.

What else will change

Pay variance is about to explode. When a single employee can do what used to require a team, compensation models built for normal distributions stop making sense. Companies that figure this out early will restructure equity and ESOP programs to attract and retain high-agency individuals. Those that don’t will watch their best people leave.

Hiring changes. The traits that mattered in a team-centric world (collaboration, consensus-building, “culture fit”) still matter, but they’re insufficient. Agency becomes the filter. Independent thinking. The willingness to build rather than wait for permission. Corporate theater won’t survive against individuals who can demonstrate outcomes in a live context window.

Methodologies won’t disappear, but they’ll change in character. The question is no longer how to coordinate humans. We’ll see new methodologies that govern how humans and agents can productively collaborate in a sociotechnical system. The rituals of Agile were designed for a world where the bottleneck was code. Now that code is abundant, the bottleneck moves to intent, to verification, to discovery.

Explore and exploit

I’ve written before about the twin engines of high-performing companies: optionality and focus. The best organizations don’t alternate between exploration and exploitation. They run both simultaneously; ringfenced bets alongside a tight, disciplined core. Optionality without focus leads to wasted potential. Focus without optionality leads to death by disruption.

AI-native velocity shifts this balance.

When execution speed increases tenfold, the cost of exploration drops. You can run experiments that used to take a quarter in a week. You can kill bad ideas faster and double down on good ones sooner. This means not taking bets becomes riskier than it was before. The companies that double down on their existing playbook without experimenting, will find themselves outflanked by competitors who are testing three new approaches for every one they’re defending.

The shadow org loses power

Every company runs on two systems. The visible one shows up in org charts, roadmaps, and OKRs. The invisible one lives in calendars, hallway conversations, and decisions that are technically reversible but socially settled.

For decades, the invisible system determined how things got done. You learned to read calendars like org charts. You figured out which documents were ceremonial and which ones mattered. You discovered who would block a decision even while nodding in the meeting.

AI changes this. Except maybe in governments and institutions but that’s another story.

When the operating model is expressed as markdown files, you can talk to it directly. You don’t need to “know the right person to invite to the pre-meeting.” When decisions are captured with reasoning attached, institutional knowledge isn’t locked in individual heads. When discussions happen with an AI in the chat, they become more factual and grounded, because the shared context is explicit.

I didn’t expect this when I started experimenting with an operating model in markdown. The speed of iteration increased, yes. But the second-order effect surprised me: company politics started melting away. The AI played a role as neutral arbiter, as long as we agreed on the shared context. Discussions that used to devolve into status games became tractable.

SOPs that existed only on paper started to matter. Not because we enforced them, but because they actively made work easier while making output more consistent. The gap between documented process and real process is shrinking.

The awkward parts

Not everyone benefits from this shift.

The employees who built careers on navigating invisible systems will find their skills devalued. The managers whose authority came from gatekeeping information will discover that agents bypass gatekeepers. The companies that confused activity with outcome will be exposed when competitors demonstrate what lean, high-agency teams can achieve.

Most organizations will do this wrong. They’ll buy AI tools without changing how decisions get made. They’ll train people on prompts without addressing the organizational drag that makes prompts necessary. They’ll automate dysfunction and wonder why they’re not seeing results.

The gap that opens in 2026 runs between companies that adopted AI properly (building discipline as a byproduct) and those that adopted superficially (accelerating waste at scale).

The trillion-dollar question

The blogosphere (well, Substack and X) are abuzz with an interesting debate: how do you build systems that make agents useful inside organizations?

The most seductive answer is “context graphs.” The idea: capture enough decision traces, enough institutional knowledge, enough reasoning, and you can build agents that understand your organization. Some version of this idea is behind nearly every enterprise AI roadmap. Jaya Gupta called it “AI’s trillion-dollar opportunity.” Animesh Akoratana proposed agents as “informed walkers” building world models from their trajectories. Kirk Marple argued we should adopt existing ontologies and focus learning on what’s novel. Gil Feig countered that the graph is the easy part; selection and coordination logic is the product. Parcadei synthesized the debate into a flywheel: ingest, store, resolve, retrieve, serve, capture, compound.

The premise of context graphs is that you can capture what employees know and encode it for agents. But can you?

Part 2 of this series synthesizes the debate and should come out in a week or so.

Part 3 offers my answer, grounded in what we’ve learned building an AI-native operating model in Voxdale (a boutique design and engineering house).

The event, combined with recent acquisition buzz around companies like Showpad (whose co-founders met at Netlog), had everyone talking about how Ghent punches above its weight in European tech.

For leaders, this concentration of ambition creates an interesting challenge: how do you keep great people engaged when they're surrounded by other fast-growing startups doing fascinating work? It's a good problem to have, but still something to solve.

This brought me back to the question I ended with in my last essay: given that any company consists of people with wildly varying quirks, needs and career ambitions, how do you get them to act in concert with the company's needs? And how do you keep talent engaged through the inevitable rough patches?

How (Not) To Talk To People At Work

I already know what doesn’t work. When I coach people who oversee teams, I like to ask them how they engage team members in individual conversations. If their one-on-ones read like a laundry list of hoops to jump through, I know I will find blind spots in their team.

Some managers ask okayish questions - “Where do you see yourself in a few years?” “What motivates you?” - but these land flat if they come across as scripted. Good conversations flow both ways. That’s easy to forget in hierarchical settings, where people take turns performing.

Most companies treat career conversations as performance theater, but they're actually your best early warning system.

I want to share with you a model for approaching conversations around career perspectives. For you to steal or give feedback on. It’s not a very fancy model. It’s just a Venn diagram if I’m being honest. But I find it helpful.

The Career Vector Map

Imagine three overlapping circles:

• Personal aspirations: who you want to become, how and where you want to live. The stuff we pretend doesn’t matter at work, until people quit because of it.

• Professional goals: what you expect from your career, what skills you want to build, what roles you’re aiming for… This is about mastery, leverage, and identity.

• Organizational objectives: what the company needs, bets it’s making, roles it’s willing to back. This is about how leaders translate vision slideware to reality.

I know 'navigating alignment' sounds like the kind of phrase that gets thrown around in recurring meetings by people who’ve not done any real work since Y2K. But strip away the jargon and that's genuinely what I'm trying to map; the literal geometry of someone's working life.

The default assumption is that alignment “just happens.” That if someone’s doing their job and getting paid, the rest falls into place. Things are rarely so simple. People and companies evolve, and their vectors fall in and out of alignment constantly.

The point of this model is to make these shifts visible. It gives people a shared surface to talk about what usually stays vague. Sometimes I actually draw the circles on a whiteboard, sometimes it stays a mental exercise. More often than not, it’s what gets the conversation off the corporate rails and into the terrain that matters.

The Permutations Are The Point

I forgive you for assuming that the goal is to get all three circles to overlap. A perfect bullseye is rare, temporary, and not always necessary.

The entrepreneurial employee

Imagine, if you will, an exceptional performer: hard-working, high-agency and passionate. How do you retain people like that? The trick is to understand what makes them tick.

People like this often dream about founding their own venture, and they may see the current engagement as training wheels. This arrangement can be mutually beneficial, even if it’s only temporary.

But you can dig deeper. How sure are they about becoming a founder? If they have what it takes, you won’t be able to stop them but the path is not for everybody... There’s a middle ground between employees and founders; tools like ESOP’s, when used wisely, allow companies to offer people a combination of skin-in-the-game and a degree of certainty about their income.

The well-liked misfit

Here’s another persona: cultural glue, morale booster, maybe part of the early team. But the role has outgrown them. Everyone avoids the topic because it feels personal.

The goal of the diagram is to make it clear when a person’s trajectories and organizational needs diverge. This helps make it not personal. It lets you talk about the delta without making it a judgment.

Then the question becomes: is there a role for someone whose professional goals aren’t evolving, but whose presence still adds value? Sometimes the answer is yes. Sometimes it’s time to say goodbye.

The fast burner

Every company has people like this: driven, ambitious, and technically sharp, but always looking for the fast track of career development. They want a leadership role, but the company’s not ready. Or they want to switch domains, but there is no vacancy in the new domain.

This is where good leaders earn their keep. They don’t say “no”, but they say things like: “not yet”, or “not here, but there”, or “not until you demonstrate your capabilities on thing X”. By definition, individual contributors have less of a bird’s eye view, and they can use help calibrating their ambitions with the available options.

These examples show how the map works for individual conversations. What most companies miss, is how these conversations should shape the system itself.

From Performance Reviews to Learning Loops

In all but the best companies, career conversations are framed like evaluations. Managers ask firm questions about progress toward goals. Employees answer. The result is a pat on the back or a warning to do better. Corporate-sponsored performance theater.

In the best companies, leadership revolves around engaging in real talk with the people in the trenches. When people tell you which interactions / processes / partners sap the joy out of their work, that’s valuable data.

Departments that ooze drama, domains that can’t seem to grow leaders, individual contributors who can predict project failures before the first letter of code is written… These are early signals that show up in people’s stories long before they show up in metrics.

Most orgs don’t have a way to catch those signals. Managers aren’t trained (or incentivized) to listen systemically. When these stories don’t even surface in private 1:1s, there’s no way to act on them.

The take-away then is to treat career conversations as a bidirectional learning loop, with the map on the table and curiosity on both sides. People should be able to take feedback, but also to dish it out.

In learning organizations, it’s the prerogative of the people in the trenches to call out their leaders on the effectiveness of their strategies and tactics. True leaders acknowledge and embrace this two-way street.

Of course, this leadership style does not work in isolation. If you want people managers with the mandate to relay messages upward, downward, and sideways, the entire organization needs to be receptive to this feedback culture.

Final Thoughts

This model is messy by design. It doesn’t spit out decisions. It requires leaders that can juggle multiple perspectives simultaneously. The pay-off is it can bring humanity back into systems that too often try to abstract people away.

Disclaimers:

• I am a proud member of the Netlog Mafia myself and I have a vested interest in propagating this brand 😎

• AI slop is taking over the Internet (“It’s not this — it’s that”). I appreciate LLMs as a tool to sharpen ideas; but I can promise my readers that I still do my own writing.

  • Here is a good essay on how to use AI without becoming stupid.

Subscribe now

“Imagine how much harder physics would be if electrons had feelings.” —Richard Feynman

Feynman meant it as a joke. But anyone who's led a team in a complex environment knows: electrons with feelings is the job description.

And yet, so much of “leadership” still assumes the opposite. We treat organizations as machines, people as inputs, and strategy as a cascading waterfall of instructions. These tactics don’t work, yet they persist.

Modern companies are sociotechnical systems: entangled networks of humans, technologies, habits, incentives, blind spots, egos and whispered Slack threads. People don’t run on logic alone. They lie, resist, adapt, and surprise. Managing people requires a different lens.

I call this sociotechnical leadership. Let’s unlearn a few things, first.

From Machines to Mycelium

The heroic leader myth dies hard. I am the first to applaud decisive moves from the top, but there is more to it. In complex systems, leadership is about shaping the conditions where emergence can happen. That’s a fancy way of saying: less commanding, more gardening.

Telling people to “be innovative” does not result in innovation. Nor does an innovation department. You get it by creating the space where unlikely collisions and informal alliances spark new ideas. You foster shared language, modular tools, and just enough slack to allow the system to rewire itself under pressure.

Think less CEO-as-commanding-officer and more CEO-as-mycologist: observing patterns, nourishing spores, letting the underground network do its thing.

The One-Track Career Lie

Here’s a quiet tragedy playing out in your company right now: your best technical talent is being slowly nudged out of the work they love because management is the only path up.

The assumption is baked into most org charts. Up means people. Down means detail. Most companies have been optimizing for the wrong kind of leverage entirely.

Not everyone who writes great code, designs elegant systems, or spots failure modes early should be herding humans in 1:1s. And not everyone who does herd humans should have to pretend they’re still “in the weeds” to stay relevant.

We need more than one ladder. Or better yet, a skill tree.

Climb Trees, Not Ladders

If you’ve ever played Skyrim or some other RPG, you know the deal. You don’t climb one ladder. You explore. You make trade-offs. You level up in different directions. Your wizard doesn’t need sword skills. Your battle knight doesn’t need to pick locks. But you need all of them on the raid.

Before you start building a career framework around this kind of map, consider this: careers never start at leadership and influence.

Agile coaches often promote the idea of becoming T-shaped —going deep in one domain while developing breadth. Some go further and talk about key-shaped profiles, with multiple verticals in parallel. Either way, mastery requires a combination of technical depth and situational awareness.

So let’s abstract that early-career exploration - whether it’s software dev, QA, design, or ops - and call it Technical Excellence. It’s the realm of focus, craft, and domain fluency.

Leadership, then, becomes a distinct progression. It’s a fork, not a level-up. Career hunters look for seniority and rewards, but the best leaders create leverage and responsibility for themselves.

This distinction matters. It solves the demoralizing issue in many orgs: the only way to “move up” is to start managing people. When you separate the paths, you make space for both kinds of excellence to thrive.

This is the idea behind the dual track career framework—a structure that lets people specialize without disappearing:

• Technical Excellence: depth, craft, autonomy, and disproportionate problem-solving.

• Leadership: coordination, influence, emotional labor, and systems choreography.

Each has progression. Each has prestige. Neither is subordinate to the other. The best orgs let people dabble, test branches and pivot. One important nuance: leadership cannot exist without a foundation of technical excellence. Leaders that become too removed from the work, stop being good leaders.

The dual track isn’t a new idea. But it’s rarely done well because it threatens the old status hierarchy. Luckily, threatening hierarchy is why this author gets up in the morning.

What Happens Next

Of course, it’s one thing to sketch a metaphor for growth. It’s another to operationalize it.

How do you actually map real people, with real aspirations and quirks, across this multidimensional terrain?

In my next essay, I’ll tackle that question.

Stay tuned.

It’s been a while… Life took me on a bit of a detour— becoming a father to my son Otto and renovating a farmhouse that I hope to call home one day. These past 18 months have been busy and fulfilling in ways I hadn’t quite imagined.

But I’ve missed writing here. Complexity and the many ways it shapes our lives, work, and decisions haven’t stopped fascinating me, and I’m looking forward to reconnecting with all of you who share this interest.

Let’s pick up where we left off. In my last essay, I explored how open-ended options and a deep understanding of non-determinism empower leaders to navigate uncertainty. However, options only add value if we can exploit them. This begs the question:

How do we Reconcile Optionality with Focus?

In business strategy, there is a constant tension between exploring new opportunities and exploiting existing ones. The former requires an openness to new ideas and a willingness to take risks, while the latter demands focus and discipline to execute efficiently.

Very few companies excel at operational excellence and innovation at the same time. Research by BCG puts the number at around 2%:

Of the 2,500 public companies we analyzed, just 2% consistently outperform their peers on both growth and profitability during good and bad times. These “2% companies,” as we call them, are able to renew themselves in large part by driving innovation and efficiency simultaneously.

Amazon has historically done a great job of pushing the boundaries in both domains:

• Exploring: Amazon consistently demonstrates its commitment to innovation with new initiatives such as Alexa, AWS, and, more recently, ventures into areas like healthcare and satellite internet (Project Kuiper). AWS, for instance, started as a risky, unconventional move for a retailer, but it's now a cornerstone of their growth strategy.

• Exploiting: Amazon’s fulfillment operations are a model of efficiency. Its use of robotics, automation, and advanced logistics has set industry benchmarks in cost efficiency and speed. This focus allows Amazon to achieve economies of scale that few can match, ensuring that core e-commerce operations remain highly profitable and defendable.

Video source

Only companies that prioritized digital transformation in the past (modern software development practices, cloud infrastructure, automation,…) have the option to capitalize on new technological waves, like machine learning. Digital laggards have closed off most technological paths. Opening them up will take a lot of time.

Optionality is the practice of keeping promising paths open. In and of itself, it does not lead to innovation or operational excellence. It only positions companies to take advantage of opportunities. To seize these opportunities, they need another critical skill: focus. By honing the ability to focus, high-performing companies learn to prioritize their efforts and exploit some of the opportunities that optionality has brought within reach.

Hence the title of this essay. To stay at the top of the food chain, 2% companies rely on the twin engines of optionality and focus. It takes both capabilities to thrive in good times and bad.

We already explored optionality in S02E02. But what does focus mean in an organizational context?

Focus = Flow

Focus refers to the ability to direct one's attention to a specific task or goal and sustain that attention over time. It is about maintaining attention to the task at hand while filtering out distractions and irrelevant information.

Focus is closely related to the concept of flow: the complete absorption and engagement in an activity. Athletes call it "being in the zone" — a state characterized by a sense of effortless concentration, a loss of self-awareness, and a feeling of immersion in one's actions.

From personal experience, we know how much we can accomplish in a flow state. We also know it’s not easy to get there. And it’s fragile—one notification on your phone, and it’s gone.

Teams can achieve a similar flow state, consistently delivering value to customers. However, as the scale of coordination grows, it becomes harder and harder to maintain flow and focus. It’s doable for small teams - this is what gives startups a leg up. In enterprises, it’s exceedingly rare.

Unlike optionality, focus requires ruthless prioritization, clear goal-setting, and disciplined adherence to metrics. This can feel constraining but is necessary for operational excellence.

Is there a management methodology that helps us internalize these requirements?

There are a few. Back in season one, I wrote about cybernetics. Any organizational model worth its salt has its roots in cybernetics. The most influential translation of cybernetics to operational practices is probably Deming’s SPC, or Statistical Process Control. It is also the hardest one to grasp by far. I may cover it in a separate essay.

In this introduction, I will focus on Kanban. If cybernetics is the trunk, SPC is the principal branch, and Kanban is an easier-to-reach sub-branch of statistical process control.

A brief Introduction to Kanban

Flow is a central concept from Kanban that tracks how value moves through the system. Most people know the Kanban board as a visual workflow management tool, but fewer understand the underlying mechanics. Let’s explore them with a familiar example: traffic.

Imagine a crowded highway with four lanes. During peak hours, the system operates near 100% capacity utilization. If one lane closes due to an accident, we see a non-linear effect on cycle time—the time it takes to travel the highway. Reducing capacity by 25% causes cycle time to rise far more than 25%. If an accident were to occur at night, it wouldn’t impact cycle time due to lower capacity utilization.

The throughput of this system is the number of cars reaching point B from point A. At night, the throughput is lower, but only because fewer cars enter the highway — not because the system is congested. The highway remains highly efficient since the few cars that do enter, travel quickly.

Applying Kanban Principles to a Software Team

If a team operates at 100% utilization, any unforeseen request (a production bug, a question from another partner in the ecosystem, unexpected discovery work,...) will wreak havoc on the cycle time of the system.

It is possible (and very common) to overwhelm the delivery part of the system if the ideation process upstream is not in sync with the throughput capacity in development. In Kanban terms: the arrival rate of items in the system is higher than the departure rate (more cars entering than exiting the highway leads to congestion).

Note: in software teams, the congestion is often caused by work being pushed into the system, rather than the team pulling work from the backlog. This video does a great job of illustrating the concept of pull:

Allowing too much work into the system slows the system down. The Kanban philosophy encourages its practitioners to visualize the workflow (including work-in-progress) and actively manage it.

If a team has 10 user stories floating between ‘in progress’ and ‘done’, WIP is 10. When WIP is high, stories spend longer waiting in the system. High WIP leads to the flow of value slowing down due to coordination overhead and context switches.

Saying that WIP is high is another way of saying that a product team has high “utilization”. Or from the team’s perspective: the team has very little slack. While not intuitive, it is necessary to leave slack in a team’s planning.

• it provides the optional capacity to absorb unexpected work, such as new feature requests or production bugs, without derailing the entire workflow. It supports optionality by keeping paths open to respond to unplanned opportunities.

• it also enables focus by reducing cognitive load, allowing teams to achieve flow and focus on fewer, higher-priority tasks. In turn, this frees up time to tackle complex problems without interruption.

Flow vs Resource Efficiency

But flow goes beyond team productivity to the organizational level. Work struggles to flow in a company organized in silos that prioritize their own interests over the customer’s. This is where resource efficiency differs from flow efficiency.

Resource efficiency maximizes the utilization of resources like people and machines, assuming the more you use them, the more output you get. This assumption only holds to a point.

Beyond a certain utilization level, cycle time and throughput decline due to congestion and queuing. Flow efficiency, however, shifts the focus from pushing the system to its limits toward enabling work to move smoothly. By focusing on flow, teams can generate more value without overburdening their resources, achieving a balance that benefits both the work and the people doing it.

To achieve flow efficiency, organizations need to organize around the value chain, not around functions or departments. This means breaking down silos and organizing teams around the flow of value to the customer.

This is a much larger topic, and later in the season, I will review techniques to address them, like Team Topologies, Wardley Mapping, and Domain Driven Design.

Bringing It All Together: Integrating Optionality and Focus

High-performing companies don’t alternate between exploration and exploitation; they balance the two, integrating optionality and focus to adapt fluidly to change without sacrificing core strengths. A focus on flow efficiency provides the flexibility to combine these engines, enabling teams to stay responsive to emerging opportunities (optionality) while remaining committed to priorities that drive value (focus).

This is the operational perspective, at least. At the strategic level, companies may place bets they deliberately compartmentalize from core operations—just as Amazon did with AWS and is doing with Project Kuiper. By keeping these initiatives separate initially, companies can explore new directions without diluting their core focus.

If the bet succeeds, these “2% companies” benefit by leveraging lessons and resources across these compartments. Two more examples illustrate this approach:

• Apple’s push into health with Apple Watch

• Meta’s bet on VR (although the jury is still out on the success of this strategy)

Optionality without focus leads to wasted potential; focus without optionality leads to rigidity. Salesforce’s acquisition of Slack illustrates the former—Slack’s integration into Salesforce has yet to materialize into growth. Intel’s loss of its edge in mass production of leading node ASICs exemplifies the latter—focusing too narrowly on established processes and missing new growth areas in semiconductor innovation.

Companies that successfully marry focus and optionality create a unique adaptability, allowing them to thrive in both stable and volatile markets.

Further reading

• Daniel Vacanti: Actionable Agile Metrics for Predictability

• Eliyahu M. Goldratt: The Goal

• Chris Miller: Chip War: The Fight for the World's Most Critical Technology

In season 2, I alternate between long-form essays and curated tidbits on organization design. In this post, I curate content from across the web.

How will AI impact the way we work? 🤖🧠

Marc Andreessen recently took to Substack. In some of his first posts, the OG venture capitalist dismissed any unemployment concerns around AI. In essence, he claims that regulatory constraints will slow down AI innovation.

On the other side of the argument, Nathan Labenz makes a strong case - I absolutely recommend reading this thread - that we are collectively underestimating the impact of AI on how we will work in the future.

Instead of slotting AI into human roles, Labenz predicts that we will break down jobs into tasks that AI can do. This is not new: it happened before with manufacturing. A century ago, making things was more like today's knowledge work. Will ML do to service work what assembly lines did to production?

Seeing what LLMs can do, I’m not betting against them. If I worked in a domain like tech support, I would start upskilling now. Of course, AI adoption will take longer in sectors that require a large degree of accuracy. For now, the lack of explainability is a showstopper for any business with low error tolerance (and there are a lot of those).

It’s time to retire the Tuckman model (with fire 🔥)

If you ever find yourself on an agile meetup wanting to stir things up, bring up the benefits of long-lived teams.

In the olden days of waterfall development, teams were far from stable as project managers selected team members from a grab bag of available people. The agile pendulum has swung completely in the other direction, sanctifying long-lived, stable teams.

This dogma is usually justified by the Tuckman Model (group dynamics of norming, forming, storming, etc..).

I want to make a case for retiring this model based on its unstable foundations and criticisms from practitioners (many such cases). I suggest a much simpler approach for deciding the optimal dynamism of your team:

• If you experience inertia and boredom on your product team, you have erred too much on the side of stability. Shake things up.

• If you experience chaos and constant relearning on your product team, things might be too dynamic. Navigate to calmer waters.

Heidi Helfand has a well-researched book on the topic for a more nuanced view.

Best kept secret of Agile Twitter 😂

Agile Twitter isn’t known for its sense of humor, and it’s about time we started changing that. I don’t know how Alyssa doesn’t have more followers so please help fix that, dear readers:

Sociotechnical research into the effectiveness of agile 🎓

I am partial to the concept of sociotechnical systems as a mental model for organization and management. It is not the only systems branch on the organizational tree, but I favor it because it combines deep scientific roots with practical applicability.

On the science side of open systems theory, Merrelyn Emery has been investigating the health of the ICT industry, with a specific focus on how organizational change affects productivity and employee well-being.

This concluding paragraph from her research is pretty damning:

After over 20 years of agile, this study shows that agile has failed to produce a coherent industry or sustainable organizations as the critical factors of high intrinsic motivation and organizational performance are in short supply. Highly desirable personal and organizational outcomes are few and far between.

I will write more on the ideas from sociotechnical systems thinking, but if you cannot wait, Trond Hjorteland does a great job of connecting the philosophy with the practice of creating software:

In season 2, I alternate between long-form essays and curated tidbits on organization design. This is an essay.

Last season, I looked at organizations through the lens of complexity theory, far removed from the day-to-day business reality. In season two, I go down a level. Instead of theorizing about evolution writ large, I’ll write about the stripes of the tiger and the shell of the turtle. Like animals, companies co-evolve with their environment, and some are better at this game than others.

In corporate evolution, rapid changes in technology are usually the forcing function. Digital transformation is the boardroom term for the evolutionary process of getting toward the desired state of business agility - the holy grail of being able to adapt to any market evolution.

In today’s article, I want to tackle the following questions:

• Why do so many digital transformations fail?

• What are the principles that underlie true business agility?

Why do digital transformations fail?

Companies are usually aware that the playing field is changing and that they need to adapt in order to survive. The trickier question is how to internalize that ability. The seemingly safe path is to hire a reputable consulting firm to come up with solutions that aren’t too much of a stretch compared to existing practices. For example:

• Erect a skunkworks circus tent and launch a digital studio that is unencumbered by the bureaucracy of the rest of the organization — this allegedly worked for Lockheed, but you are more likely to subsidize rival factions. The IT department down in the basement won’t be fond of the hipster studio.

• Adopt the latest trending architecture (eg MACH) — not featured in the consultant’s slides: the social and technical challenges hidden underneath this recommendation

• Implement a scaled agile framework — no one ever got fired for recommending SAFe, but I don’t know of many successful cases either.

When companies hire management consultants, it is because they need to get something complex done (and things get complex quickly in enterprises). To persuade others to adopt a thing, enterprise execs usually need more than logic; they need evidence that can withstand scrutiny and criticism when someone inevitably questions the decision.

Enterprise companies turn to the McKinsey’s of this world to rationalize and clarify their decisions. They’re buying a seal of approval, not original thinking. There is nothing inherently wrong with this approach, but it does lead to mimetic behavior.

Take agile software development. Over the past decade, enterprises have been flocking to agile as an organizational fountain of youth. If that sounds too good to be true, it is. Any moment now, companies will start to realize that their expensive agile software development frameworkᵀᴹ did not translate to actual business agility.

The failure of digital transformation programs begins with outdated mindsets at the board and C-suite levels. Instead of addressing this behavioral change from first principles, many companies attempt to purchase a packaged solution.

Agility is not something you can buy off the shelf, but there are plenty of agileᵀᴹ certified consultants who are willing to take your money and claim otherwise. This usually results in cargo-cult agile - the superficial adoption of agile practices without a deep understanding of the underlying principles.

I’m not sure if I’m still allowed to define a cargo cult in 2023, so I have asked an LLM to do it for me:

Imagine you live on a remote island in the Pacific Ocean. One day, you see strange metal birds flying over your head. They land on a nearby airstrip that was built by foreign soldiers during a war. Out of these birds come men in uniforms who bring food, medicine, tools and other goods. You are amazed by these gifts and want to receive more of them.

You observe how the soldiers behave: they wear headphones, wave flags, march in formation and salute each other. You think that these actions must be some kind of magic that summons the metal birds and their cargo. You decide to copy them: you carve wooden headphones, make flags out of leaves, march around and salute each other.

You hope that by doing this, you will attract more metal birds and more cargo. But nothing happens. The war ends and the soldiers leave. The metal birds stop coming. You are confused and disappointed.

You have just become part of a cargo cult.

Halfbaked transformation initiatives lead to frustration and disillusionment with the agile philosophy, further entrenching traditional development practices. In a failed attempt to one-click-install agileᵀᴹ, the enterprise ends up even further removed from its stated goal of business agility.

First principles beat imitation

Modern digital leaders understand that there are no turnkey solutions when it comes to organization design and business strategy. Instead of imitating patterns, they look beyond the boundaries of the current best practice.

A world without best practices is a scary vision for traditional managers, but there are ways to deal with variation and non-determinism.

True leaders articulate the principles that underlie their views on management and update their priors with new information. I believe that this Bayesian approach is a managerial superpower. I will come back on probabilistic methods throughout the season but here is a short primer:

What are the first principles of business agility?

Optionality is one way to use Bayesian reasoning to make better decisions in volatile and uncertain environments. Nassim Nicholas Taleb popularized this concept in his book Antifragile. Taleb defines optionality as "the property of asymmetric upside (preferably unlimited) with limited downside":

Taleb arrived at optionality from the perspective of financial options trading. In the world of capital allocation, the concept seems to come naturally. Venture capitalists or private equity firms face uncertainty about the future performance of investments, and they leverage optionality to hedge their bets:

• Develop models to find the companies with low downside risk but high upside potential.

• Holding cash as a reserve to take advantage of opportunities as they arise.

• Diversifying across sectors, geographies, and stages of development.

In today’s markets, no matter how much analysis we perform, we rarely know with certainty what the best course of action is. In any complex adaptive system, open-ended options are an excellent alternative to knowledge and prediction. To leverage the concept of optionality, we need to maximize the options at our disposal. Options give us the right - but not the obligation - to take an action.

Companies are not that different from capital allocators. They too have finite resources to allocate and strategic bets to place. Great leaders manage their company as a portfolio of real options, essentially creating a probabilistic framework for decision-making that allows them to adjust their strategies and actions based on new information.

How does optionality enable business agility?

Leaders who understand the underlying principles of optionality, non-determinism, and complexity will have a much stronger grasp on the implementation of lean management and agile software development.

For first principle thinkers, lean and agile are not a matter of imitation. They are a matter of minimizing waste and maximizing value. Truly agile companies invest only in what is necessary while building in the flexibility to pivot if market conditions or customer needs change.

They appreciate that the lowest common denominator implementation of agileᵀᴹ (JIRA, sprints, backlogs, story points,…) does not automatically unlock the ability to pivot or jump on an opportunity when one presents itself.

Instead, modern leaders leverage their understanding of the underlying principles toward true business agility.

Let’s make this tangible.

Aligned autonomy

Product teams can only adapt to changing market circumstances when they have the autonomy to do so. However, their hands are often tied in an enterprise context:

• In hiring and rewarding people, a product team is constrained by an HR department.

• In selecting tools and partners, they are constrained by a procurement department.

• In deploying software, they are constrained by the enterprise architecture.

Autonomy is not a goal in itself. The above constraints are only a problem when the different parts of the organization are not aligned. If the product team is optimizing for agility (“we need a time & materials agreement with partner X”) but the procurement team is optimizing for control (“we want fixed scope and budget”), the company is wrestling with itself.

Top-down or bottom-up transformation?

This is a false dilemma.

In the procurement vs product team example, leadership has a role to play in setting the direction but the lack of alignment may just as well be identified from an operational point of view.

If the boots on the ground are permanently at odds with the generals, you should fix that problem before attempting a digital transformation. Let’s assume a minimally functional enterprise.

Leadership must have an excellent grasp of complexity theory and its downstream ideas of Bayesian reasoning, optionality, agility… There are countless techniques in those toolboxes but they don’t come prepackaged. Every context is unique, and leaders will have to do the hard work of figuring out what makes sense for their company.

At the same time, there is plenty to tackle on the operational level:

• Take behavior-driven development (BDD) and test-driven development (TDD), for example. These practices create the option to change the software by telling us when we have broken it. The earlier we catch errors, the easier it is to change course.

• The same can be said of the practice of ensemble programming. The best way to learn as a team is to actually collaborate and get the most out of each other’s unique set of skills, perspectives, and experiences. Teams that work together like a jazz band benefit from immediate feedback, faster flow, increased focus, and better software quality.

Companies are socio-technical systems

When we are talking about alignment and collaboration, these are social challenges. At the end of season one, I introduced the idea of socio-technical systems thinking, an approach that recognizes the interaction between people and technology in our workplaces.

The socio-technical approach to architecture is a good example. It emphasizes modularity as an important design principle for agility. Sociotechnical architects strive to align architectural boundaries with team topologies, domain knowledge, and business value. Good architecture is about more than producing UML diagrams in Archimate.

Subscribe now

Limit your downside

Optionality (and downstream ideas like lean and agile) is not just about capturing the upside. They can also help in managing risk. To do that, teams and companies want to avoid decisions with a fixed upside but an unlimited downside. For example:

• Accruing technical debt. On the upside, you get to release slightly sooner, but on the downside, you are jeopardizing the ability of the product to evolve in the long term.

• Postponing infosec investments. On the upside, you get to cut some costs or avoid those pesky ISO27001 audits. On the downside, you might find yourself locked out from a prospect’s RFP. Or much worse: locked out of your data and being held hostage by ransomware attackers.

The best options are asymmetrical in the other direction. They have a fixed cost but an unbounded upside. John Cutler has a great visualization that looks at optionality from a capacity planning perspective:

It is also a good introduction to the logical follow-up questions. Companies have to explore and exploit. At what point do we sink our teeth into something? And how do we balance optionality and focus? I’ll come back to these questions in my next essay.

Event: Teal Around The World 🤝

If you have even a passing interest in organization theory, you will have heard of Frédéric Laloux. With his famous book from 2014, he was one of the first people to capture the zeitgeist and make a case for reinventing our organizations.

With the Teal Network, a community emerged to further this mission. Next week, they will host a global conference with Frédéric Laloux as the keynote speaker.

I will be attending the event on Friday, March 3rd, and I hope to meet some of my Belgian readers there. Click here to register, too.

For those outside Belgium, find out more about the global event here.

Architecting spaghetti 🍝

Eduardo Da Silva end Nick Tune wrote an excellent article on a challenge that will ring familiar to many people in IT: when and how to modernize spaghetti code monolith architectures.

This article builds on some of the patterns in Team Topologies, the book on every CIO’s nightstand. I will cover this approach in more detail in the course of season two.

Quote of the week 💬

Redesigning the lines and boxes in an organizational chart is simpler than changing how that organization in fact operates. Changing the rules and regulations is simpler than eliciting behavior that conforms to them.

—James C. Scott, Seeing Like a State

Psychological safety 🦺

In all companies, the most wicked problems come from the issues that people are hesitant to bring up. Dancing around a problem won’t help you fix it.

Psychological safety became MBA-famous after Google labeled it the driving force for team performance.

HBR wrote a decent introduction to the concept. However, they underemphasize the impact of groupthink on psychological safety. I’ll address that intersection in the course of season two.

The machine learning revolution 🤖🧠

If you hadn’t noticed, the ML paradigm shift has officially gone mainstream:

It’s hard to keep up with everything going on in this space, and I compiled some sources for your convenience:

• When it comes to digital strategy, there are two Ben’s who offer reliably insightful analysis:

  • When Stratechery’s Ben Thompson refers to something as the most surprising and mind-blowing computer experience of my life, you want to be paying attention.

  • Every February, I look forward to Ben Evans’ yearly presentation of macro and strategic trends in the tech industry. This year’s burning question: what is the right abstraction to understand generative machine learning? What does it mean for search, product, and how we build software?

• If you want to try to ingest everything that is going on in the space (reader, beware), I can recommend another Ben who is doing a great job of curating the space through a daily newsletter: Ben’s bites.

• If you’re not following Ethan Mollick on Twitter, you are missing out on some very cool prompt engineering and analysis.

In trying to keep tabs on the capabilities of the machines, it’s worth remembering that digital transformation is upstream of AI risks and opportunities. If your organization is still operating under the industrial paradigm, it needs to fix that problem first.

I am going to make some changes to the newsletter format and frequency. The combination of weekly long-form content and curated links is a bit much. I suspect this is true both for the writer and the reader. Starting next week, I will alternate between curation and essays.

Today, I wrap up this first season. While I have only scraped the surface of the body of knowledge that is complexity theory, I feel that I have done enough to explain why we should question our default approach to organization. At the end of this episode, I introduce what’s in store for season two.

“No problem stays solved in a dynamic environment.”

—Russell L. Ackoff

S01E15 A phase transition is brewing

In last week’s episode, I investigated how complex systems emerge from simpler systems at the edge of chaos. It became clear that self-organization is a fundamental feature of complex, adaptive systems.

Self-organization and interaction with the environment allow systems to adapt and evolve. In a reinforcing feedback loop, the newly emerged structure increases the scope of cooperation and competition, enabling yet more self-organization and evolution:

Examples of this feedback loop can be found in the natural sciences (physics, biology, astronomy…) and social sciences (psychology, economics, sociology…). I’ll give examples from different domains, and try to shed some light on the enigmatic link between self-organization and emergence - the phase transition.

In early human history, when hunter-gatherers abandoned their nomadic existence to self-organize into settlements, this set off a phase transition. New societal structures started to emerge:

• increased food production and food surpluses led to larger populations

• larger populations gave rise to new social hierarchies and classes

• social technologies (specialization of labor, law, writing…) enabled new ways to compete and collaborate

• social complexification led to the invention of new tools and technologies (irrigation, pottery, metalworking…)

The agricultural revolution introduced new non-zero-sum games and heralded an increasingly sophisticated society. While there have been historical periods of decline (e.g., after the fall of the Roman Empire), it is fair to say that directionally we are evolving toward more complex societies.

Agriculture was the first of many societal phase transitions. Historians usually refer to them as revolutions, like the scientific, industrial, urbanization, and information revolutions.

What do we know about phase transitions?

Phase transitions are an endless source of fascination for scientists. Entirely new properties spring to life when systems transition from one phase to another (e.g. from ice to water, or from nomadic culture to agriculture). Much remains to be discovered - understanding these evolutions is one of the fundamental questions that complexity theory addresses.

Some patterns do emerge. Repeated perturbations in a system - the result of the conflicting forces of competition and collaboration - result in a “frustrated state” and cause increasing instability in the system’s equilibrium. Instability impacts the anchors of that system and sets off a phase transition until a new equilibrium is achieved.

In thermodynamics, like in the liquid-gas phase change of boiling water, these transitions are relatively well understood (or so my physicist friends assure me). Our understanding of these transitions becomes fuzzier when we explore human and societal interactions, as do the boundaries between states.

Phase transitions can be observed at different levels of analysis (organic, molecular, organic, psychological, social…). For example, we’ve all witnessed the phase change when laughter suddenly lightens the mood in a roomful of people. We can see them after an outstanding music performance when applause transitions into a standing ovation.

I started my career in traditional project management, and I have been at the center of more than one phase transition of truth. This phenomenon occurs when the project indicators in a spreadsheet suddenly go from orange to red, catching managers off guard (but rarely the developers).

This is an interesting aspect of phase transitions. The emerging changes often aren’t visible on the surface. In an example from physics, when ice melts it reaches a plateau where heat is continually added to the system, but it takes time for the phase transition to reach the liquid state.

The above image might trigger flashbacks to physics classes. In fact, there is a metaphysical question I have to address at this point. One of the most fundamental laws of physics states that in closed systems, the total entropy increases over time:

The law that entropy increases—the Second Law of Thermodynamics—holds, I think, the supreme position among the laws of Nature.

—Sir Arthur Stanley Eddington

How, then, can we reconcile increasing disorder (entropy) with increasing sophistication (complexity)?

I will address this question with my trademark weapons: clarification through metaphors, triangulation of sources smarter than me, and a well-chosen Youtube video. I would also like to throw down the gauntlet to my more math- and physics-oriented friend Maarten Mortier in the hope that he takes a stab at the topic.

What is entropy, and why does it matter?

Entropy refers to the natural tendency of things to lose order over time. It is the reason why abandoned houses disintegrate into the environment, why cars rust, and why software systems tend to evolve into big balls of mud.

In theory, with infinite monkeys hitting keys on typewriters, one of them could reproduce the collected works of Shakespeare. In our entropic reality, the chances are vanishingly small. The second law is a matter of probability - there are many disordered states and few ordered ones.

Entropy is a scary subject for non-physicists. For an eye-opening, yet accessible introduction to entropy, emergence, and complexity, I warmly recommend reading Sean Carroll’s book The Big Picture: On the Origins of Life, Meaning, and the Universe Itself.

Sean Carroll is a professor at Johns Hopkins and the Santa Fe Institute (the mecca of complexity science). In his book, Carroll sheds light on the surprising role of entropy as the driving force behind complexity. While this sounds like a paradox, he explains how entropy, the very concept that will eventually lead to a boring and lifeless universe, is a necessary requirement for emerging complexity:

While there is no escaping the second law of thermodynamics, we can fight back against entropy. In order to successfully curb entropy and create stability, structure, and simplicity, we need to expend energy. Order requires effort.

Physicist Sharon Glotzer has conducted groundbreaking research on this topic, specifically on the principles that govern how macroscopic properties emerge from microscopic interactions. In exploring this question, she connected entropy with self-organization:

“We typically think entropy means disorder, and so a disordered structure would have more entropy than an ordered structure. That can be true under certain circumstances, but it’s not always true, and in these cases, it’s not. I prefer to think of entropy as related to options: The more options a system of particles has to arrange itself, the higher the entropy. In certain circumstances, it’s possible for a system to have more options – more possible arrangements – of its building blocks if the system is ordered.”

— Sharon Glotzer in Wired

She further explains that entropy sets particles to “wiggling.” As it turns out, particles naturally maximize their individual space for movement. On their entropic quest for wiggle room, particles order themselves into more complex - and more ordered - states.

This is a good moment to reassure my esteemed readers. I know that you signed up for digital transformation wisdom, not for the wiggling of theoretical particles. I will not venture further on the ice of complexity (lest I fall through it), but I felt the need to show the reader that complexity theory is on solid scientific footing.

Entropy may be an abstract concept, but organizations deal with it all the time. Understanding concepts like entropy, even at a high level, can help inform very practical decisions in business strategy or software architecture.

Disorder and variation form the surface area for innovation. The companies that successfully wrangle entropy will outmaneuver the companies that merely follow a script, incapable of trying new things (see: live players vs dead players).

What’s in store for season two?

With the theoretical groundwork in place, I want to explore the ‘how’ in more detail. How can we apply complexity theory in management practice?

Over the next dozen or so episodes, I will explore the expansive and varied toolbox of sociotechnical systems management. There are many useful techniques, tools, and frameworks out there. We just need to put in the effort of learning how and when to use them.

Sociotechnical?

So far, I’ve referred to organizations as complex, adaptive systems. As we move from theory to practice, it is more helpful to think of companies as sociotechnical systems. In the words of Dave Snowden: people aren’t ants.

The sociotechnical frame implies that organizational systems can only be understood and improved if both social and technical aspects are considered and treated as interdependent parts of a complex system. Most approaches to digital transformation start from an overly narrow perspective (most often technological).

The sociotechnical systems approach is rooted in complexity theory and systems thinking - it takes a broad view. Concepts like feedback, modeling, self-criticality, emergence, and entropy can help account for the social and technical drivers of an organization (as well as the interconnectedness between its social and technical attributes).

What to take away from season one?

For all this writing on complexity, the conclusion is a simple one: we are in the midst of a phase transition. Organizational systems have entered into a frustrated state.

Under a thin veneer of engagement, most employees are larping their jobs. Dilbert cartoons are funny because they are true. Actual motivation, cooperation, or innovation are scarce in the corporate universe (exceptions exist, there is an avant-garde cohort that is taking a different approach).

I made the point that most organizations don’t live up to their potential because they run deprecated managerial software. Traditional operating models are a legacy from the industrial era. We have been trying to manage complex systems through convoluted control structures, rooted in a mechanistic worldview.

I believe that complexity theory contains the seeds for a new management theory, one that enables sociotechnical systems to thrive in the variety of the information era. Instead of focusing on control and uniformity, companies will learn to embrace flux, change, and the forming and dissolving of patterns.

Reductionist methodologies, checklists, and generic scripts will give way to the science of uncertainty. The new tenets are:

• Take more, but smaller steps. This allows organizations to:

  • experiment

  • learn faster

  • minimize risks

  • reason about cause and effects

• Normalize change and think probabilistically. Taking a linear approach to causality is futile: even if we know all the inputs, we cannot predict how a system will behave.

  • Improve your understanding of the system through mapping or modeling, but appreciate that the system reacts to the model.

  • Focus not just on the components but also on the relations between the components (both social and technical).

• Create structures that let people work their magic. The recent leap forward of AI has reaffirmed Clarke’s third law: sufficiently advanced technology is indistinguishable from magic. How to make it as a leader in the magical era?

  • Do not try design outcomes. Instead, reason about the rules, habits, and incentives that we allow to shape the system (“governance”).

  • Systems change over time, and change is driven by numerous iterations of very simple rules. To change the behavior, start with the rules.

  • Treat employees as wizards, not as serfs.

🤓 Further reading

The Big Picture: On the Origins of Life, Meaning, and the Universe Itself - Sean Carroll

Hopefully, this has resulted in an appreciation for the unpredictable web of interactions, constraints, and incentives in complex adaptive systems. At a high level, we understand that countless feedback loops can eventually lead to emergence, but we haven’t yet explored this step in detail.

How and why do complex systems emerge from simpler systems? And what does that mean for organization theory and management practice?

Physicists propose that complexity must evolve from a delicate balance between order and chaos, a state known as self-organized criticality (or more colloquially known as the edge of chaos). In this Goldilocks phase (from the fairytale where the porridge temperature needed to be just right), the components of the system are neither inert nor turbulent, but rather exist in a state of dynamic equilibrium.

In simpler terms: complex life needs a sweet spot to flourish. In this phase transition zone called criticality, there is sufficient stability to sustain life while also providing the creative spark necessary for evolution. Systems between order and chaos have evolutionary advantages with respect to ordered and disordered systems.

Look at nature. What happens when a natural system, like a forest, is too ordered? In monocultures, all trees are the same species and age. These forests are always artificial. They simply don’t evolve in nature because a monoculture is less resilient to disturbances or diseases, and less able to support a diverse range of life.

On the other hand, when there is too much chaos (wildfires, flooding, volcanic eruptions,…), complexity and life cannot emerge either. If grazers on the savanna were to move in disorganized and chaotic patterns instead of in herds, they would quickly be overwhelmed by predators.

The concept of criticality can also be observed outside of natural systems. In fact, we can simulate it on computers, thanks to John Von Neumann (whose brilliant mind we can also credit with the invention of the foundational architecture of modern computers).

In the 1940s, Von Neumann studied cellular automata. These are simple algorithms consisting of 2 elements:

• a grid of cells, each of which can be in one of a finite number of states, and

• a set of rules that determine how the states of the cells change over time.

Most of these algorithms result in boring behavior; they either die out or settle in repeating patterns. Some, like the automaton in the above gif, result in interesting, complex behavior. The rules that exhibit the most complex and interesting behavior (such as the famous "Game of Life" rule) reside on the edge of chaos.

If we can recreate lifelike behavior on a computer while also reproducing this Goldilocks zone of emergent complexity, it is not a stretch to claim that the principles of criticality are a fundamental feature of complex systems across different domains.

Besides criticality, what other circumstances do we need for complex, adaptive systems to emerge?

In the introduction (and in episode 5), I mentioned that complex adaptive systems are capable of information processing. These are 2 rather abstract concepts and it’s worth pausing at how that works. Let’s revisit these cellular automata to help us understand how complex adaptive systems compute.

Imagine that our cellular automaton operates in a grid (as above), but instead of colors, each square can be a number or a letter. We also know that the decision rules for the changing patterns can involve math or logic. This way, by letting the game run for many turns, the grid of squares can perform calculations and solve problems.

Some cellular automata (like Rule 110) have even proven to be Turing-complete, meaning they are capable of universal computing. Given enough time and memory, these simple automata can solve any problem that can be represented in a formal language.

Natural systems are also capable of computing. Take ants as an example. While there is no central authority, the colony as a whole functions as a decentralized network of millions of autonomous ants, each of which makes decisions and takes actions based on a limited set of interactions with other ants.

Ant colony computation is distinct from that of traditional computers. Ants cannot rely on the Von Neumann architecture (with a central processing unit and random-access memory), but they get the computational job done nonetheless.

To what purpose, you ask?

Computation is simply what a complex system does with information in order to adapt to its environment. Back in episode 4, I gave some examples of the evolutionary purposes of computation in ant colonies:

• scaffolding

• defense mechanisms

• predatory behavior

As long as the system operates in the zone of criticality, a few very simple rules on the component level can enable complex behavior and computation at the system level.

But we are missing another part of the emergence puzzle. How do systems in nature arrive at this Goldilocks state if no one is tuning the system from the outside?

The answer lies in self-organization. Complex systems achieve criticality from the bottom up, through the interactions of the parts of the system and their environment. These interactions aren’t random. To drive the evolution of hierarchical complexity, components interact through a combination of collaboration and competition:

• Competition over limited resources (food, territory, mating partners…) can lead to the evolution of strategies that help agents survive and reproduce in their environment. It can also lead to the emergence of hierarchies or dominance structures within a population.

• Collaboration - working together to achieve a common goal - can lead to the evolution of cooperative behaviors, such as altruism and specialization of labor.

As the components interact and compete, they often find themselves in a state of tension and unrest, leading to a breaking point where self-organized criticality occurs, like a phase change. This kicks off a transition to a new and more integrated state, marked by increased complexity.

Complexity scientists are uncovering evidence of these dynamics between order and disorder in many domains:

• One of the oldest examples supporting the self-organized criticality theory comes from biology, where we know single-celled organisms can come together to form a multi-cellular organism.

• In the economy, we see people organizing in companies, and we know that this competition and collaboration leads to the emergence of new industries and market structures.

• There exists a concept in evolutionary biology called the Red Queen hypothesis. This is the idea that organisms must keep evolving in order to keep up with the changing environment and other organisms, as if they were running on a treadmill. It's often used to explain why sexual reproduction persists.

• Self-criticality is a hot research topic in neuroscience. One interesting field of study is exploring whether consciousness can be explained as a phase transition between order and chaos in the brain. When a sedative knocks out our consciousness, that could be the drug nudging our brain slightly toward the ordered zone. On the other hand, there is increasing attention to the potentially favorable effects of psychedelics on our creative capacity. This increased creativity may be explained by the drugs moving our brain activity toward the chaotic zone.

Apparently, self-organized criticality, computation, and emergence lead to evolution. Complex systems have the ability to accumulate experience through survival, sexual selection, and/or learning. Even the most entrenched living systems will eventually fall to the relentless march of progress, constantly revising and rearranging their building blocks as they gain experience.

Through a series of hierarchical emergences—a nested sequence of parts coming together to form ever-greater wholes—the universe is undergoing a majestic self-organizing process. In other words, nature’s simplest parts organize themselves into wholes, which become the building blocks for the next level of complexity.

— Bobby Azarian (author of The Romance of Reality)

While this has interesting philosophical and spiritual implications (the meaning of the universe is… life and complexification), my mission focuses on the organizational implications.

So what should we take away as we shape our organizations and institutions? Complexity theory tells us that systems self-organize toward a zone between order and chaos, where they compute survival strategies through collaboration and competition.

I’ll start with the most evident conclusions (evident to most readers, maybe not so much to those sitting at the wheel of institutions and enterprises).

Organizations that aren’t conducive to self-organization are destined to slip into a state of dormancy (erring too much on the side of order).

How can we recognize overly ordered organizations and limits on self-organization?

• A lack of flexibility in decision-making processes

• A rigid chain of command

• Strategy translates to execution through heavily analyzed, multi-year roadmaps

• Bureaucratic processes slow down decision-making and impede progress and learning

• Employees are seen as small, interchangeable parts in a larger system

Why is this a bad thing?

• Limited ability to respond to changes. Overreliance on a formalized system of rules and policies will hinder decision-making at the edges. Ordered organizations are good at exploiting (as long as the environment is stable) but not at exploring. Order and innovation don’t mix.

• Inhibition of institutional learning. Organizational systems and their environments are always in flux. This limits the horizon of predictability. Organizations that promote the illusion of certainty will lose out to companies that take a probabilistic, Bayesian approach to planning and learning. Fast and lean experimentation beats slow, risk-avoiding analysis.

• Low employee engagement & high attrition. When organizations treat employees as interchangeable cogs in the machine, these people will go through the motions without engaging with the organizational mission. In order to get people to act as owners, they require a sense of purpose and autonomy (at the very least).

Note that enabling self-organization is not the same as eliminating all hierarchy or leadership (this will be the topic of a future episode).

On the other hand, there can be organizations where chaos reigns. While not unheard of, disorder is more easily recognized, and it tends to be a temporary state.

There are also less evident conclusions, that I will tackle in more depth in future episodes:

• There is a central role for conflict in organizations that embrace change and self-organizing criticality.

• Systems change over time, and change is driven by numerous iterations of very simple rules.

• Competition and collaboration lie at the heart of evolution. The best leaders can reframe challenges and transform them into non-zero-sum games.

Companies that succeed in surfing the edge of chaos will see themselves rewarded with truly agile characteristics, like rapid information processing, collective responsiveness to perturbations, and the capacity to incorporate a vast array of external stimuli without reaching saturation.

Companies that miss the edge will succumb to creative destruction and join the statistical rank of shrinking company lifespans.

Share

Best of the Rest 👀

🤖 Software AI is eating the world

There is little doubt that generative AI will create massive disruption for all knowledge work. The only questions are how and when?

• A16z tries to answer the question of where value will accrue and asks whether AI apps need to own the model to guarantee differentiation and retention.

• One emerging approach is a domain model, where companies take a narrow, vertical approach like this platform for AI-generated gaming assets.

  AI’s application layer

• The Digital Native Substack also has a good, broader take on the application landscape of generative AI.

🤝 How we organize

• If you spent some time learning about incentives and goals in business, you’ll have heard of Goodhart’s law: '“when a measure becomes a target, it ceases to be a good measure”. Cedric Chin of Commoncog wrote an excellent piece on doing something actually useful with this law.

• The convergence between IT (information tech) and OT (operational technology) is inevitable. Manufacturing organizations have resisted change for a long time but are starting to realize that robustness and adaptivity are not mutually exclusive. Jan Bosch has started an insightful series on product development fallacies in manufacturing & industry (Read them as LinkedIn posts, or on Jan’s blog)

• Many companies are still struggling with the fact that agile methodologies do not always equate with agility. I don’t agree with everything in this article equating Scrum with SAFe, but overreliance on methodologies can lead to teams running in circles around a local optimum.

• Alex Ewerlöf describes the wonderful things that can happen when teams start dedicating fixed bandwidth to the removal of tech debt.

🤓 Further reading

• If you are interested in how this process of complexification translates to meaning, I can recommend this book by Bobby Azarian or this excellent podcast of John Vervaeke with Lex Fridman.

• I found inspiration for this episode in many places, but one source was especially insightful:

Intimations of a New Worldview

Intimations of a New Worldview

Introduction This is going to be a long one. It is also pretty technical in places, although I have tried to keep the jargon to a minimum. In this case I think the length and technicality are worth it. The title of both this essay and the substack I’m publishing it on is “Intimations of a New Worldview.” This post is an outline of what I believe that wor…

Read more

4 years ago · 30 likes · 8 comments · Brett Andersen

Subscribe now

Last week we left off with Formula One, which is mostly a technical system. Although the drivers do play a role, teams select them expressly to minimize the human factor. While some drivers may be able to ride the occasional perfect race, the championship-winning teams prioritize consistency over flash. The perfect driver is a human metronome.

In this publication, we are interested in systems where humans play a bigger role. Every organization is an open & interactive system populated by agents that are notorious for having free will. When free will, evolution, and learning are involved, prediction becomes difficult. In social systems, agents predict and react to the actions and predictions of other agents in their environment.

The above is a description from the outside in, but it can be helpful to consider this from our own perspective. Tim Urban has an eye-opening visualization:

Now, imagine modeling a social system full of sentient agents, learning, adapting, evolving, and vying to optimize their life paths.

Whatever the perspective, social systems quickly become nonlinear and difficult to model. Even if we have a solid understanding of all of its parts and their interactions, deterministic prediction is impossible because of sensitivity to initial conditions (and other aspects of chaos theory that I won’t go into lest I stumble into a new rabbit hole).

“Imagine how much harder physics would be if electrons had feelings.”

— Richard Feynman (presumably)

While it is difficult to accurately predict the outcomes or behaviors of a social system, that doesn’t mean models are useless here. The domain of simulating the interactions of autonomous agents is called agent-based modeling, and this approach is gaining popularity in the social sciences.

Since we are more interested their organizational application, I went looking for some illustrations of agent-based modeling in a business context. The best examples I have found, are from the crypto industry. We’ll need some context about this business, first.

“Only when the tide goes out, do you discover who’s been swimming naked.”

— Warren Buffet

Because crypto innovation is both permissionless and tightly coupled with speculation, its tide ebbs and flows more spectacularly than that of traditional finance (Buyer, beware!). People who follow the industry over several cycles learn that when the tide recedes, there is more to be observed than people swimming naked.

During bull markets, it is very hard to distinguish valuable and authentic projects from the swindlers and the hype. In crypto winters (and we are in one now), it becomes clear which primitives have staying power. I want to zoom in on such a primitive: stablecoins. They are one of the foundational innovations in decentralized finance, with volumes rivaling those of traditional card networks.

What are stablecoins, exactly?

Stablecoins are programmable, fungible tokens that are designed to have a stable price. They came into existence as a solution to address the volatility associated with other cryptocurrencies. Price stability aside, the value proposition of stablecoins is not all that different from other cryptocurrencies (trust-minimized, programmable, censorship-resistant currencies).

Some designs have proven more stable than others, and I’ll explore a stablecoin design that has been around for a while: MakerDAO’s Dai. So how does Dai work? How would we model a stablecoin?

Dai is able to maintain its stability through the use of collateralized debt: in order to take out Dai, you first have to deposit collateral in the form of digital assets. This logic lives in smart contracts on the Ethereum blockchain, and this contract platform is managed by the decentralized autonomous organization called MakerDAO (I’ll come back on DAOs in a later episode that tackles innovation in organization structures).

What does any of this have to do with complex, adaptive systems?

Well, the Maker platform collectively manages the supply of Dai to ensure its stability. Each person in this system is an agent, with their own unique goals and behaviors. This can be a crypto investor anticipating a downturn, or it can be a remittance use case, where someone wants to move money across borders quickly and cheaply.

Whatever the reason for wanting to acquire Dai, the user must first provide collateral in the form of another cryptocurrency, such as Ethereum. They are then able to mint Dai. However, if the value of the collateral drops below a certain threshold, they will have to provide more collateral or have their loan liquidated. This creates an incentive for agents to be careful about the types of collateral they provide and how much they borrow.

The MakerDAO members themselves are important agents in the system. Incentivized through the value of MKR (a separate token from Dai), they oversee the functioning of the system and make decisions on how to maintain the stability of Dai. They can decide to increase or decrease the amount of collateral required to create new Dai, or they can change the interest rate on the debt to encourage or discourage people from creating new Dai.

Through the interactions of these agents, the supply of Dai is automatically adjusted to keep its value stable. If the demand for Dai increases, the price will rise, and more agents will be incentivized to provide collateral to borrow Dai, increasing the supply and bringing the price back down. If the demand for Dai decreases, the opposite will happen.

Here is a visualization of these mechanics in a diagram:

Modeling is closely related to control theory, and the above diagram uses control concepts (sensor, actuator, controller,…) to illustrate the design of this stablecoin. The practice of modeling is actually well-established in the crypto industry, probably because of its permissionless and open nature.

Another modeling illustration from DeFi (decentralized finance) is Uniswap; a decentralized exchange where people can trade tokens. Uniswap uses a market-maker formula to match buyers and sellers. This is a type of control system that uses mathematical equations to balance supply and demand in the market. This model lets you look under the hood of Uniswap’s market maker.

The above examples are regulation systems. Their ability to successfully regulate a complex adaptive system is wholly dependent on whether the model is a realistic representation of reality. Do Kwon, the founder of the now-defunct Terra stablecoin, also had a model. It just wasn’t a very successful model because it failed to capture the behavior of the real world (apparently, you cannot back a stablecoin by a sister coin).

Just as the sculptor chips away the unnecessary marble to reveal the masterpiece, a modeler of complex systems must filter out the noise to reveal the underlying structure. It requires both art and craft to create a representation that we recognize as valuable.

Good models (like the MakerDAO and Uniswap examples) demonstrate how system-level properties emerge from individual behavior. They help us understand and possibly regulate the system.

Does this mean that agent-based models can help us predict the behavior of complex systems?

The answer is nuanced. It is possible to use agent-based models to simulate the interactions of individual agents in a system. This can tell us something about future behavior as long as the model is based on many different types of local behavior (as opposed to the mechanistic approach of relying on statistical averages for the behavior of many individuals).

While useful for tokenomics designers, agent-based models won’t predict black swan events. At best, the modeler develops an intuition for how different types of agent-level interactions can lead to emergent patterns and behaviors. That’s not quite the same as predicting the future. Many a sorcerer’s apprentice has learned this lesson the hard way.

Herein lies the key lesson of modeling complex systems. They can help us to better understand the underlying probabilistic nature of complex systems. Unlike deterministic models, which assume that a particular outcome is certain given a set of initial conditions, we take into account the inherent uncertainty and variability in the behavior of agents.

In the upcoming season, I will zoom in on digital transformation, and give some concrete examples of how organizations can get a better handle on the future by understanding variation and by using probabilistic approaches toward planning (strategic and operational).

But first, I’ll wrap up this season and further map the intersection between complexity and organization theory.

Best of the Rest 👀

🤖 Software AI is eating the world

• The 4th age of programming is here. What will be the 2nd order effects as developers increase their productivity and scope through AI?

  • Will teams remain the smallest value-adding unit? There is more value in teams than just the sum of their skills, but it's likely that teams will become smaller.

  • Team-level agility will get easier, and the focus of the agile "industry" will shift from the team to the coordination level. Architecture, culture, and governance will become the drivers of organizational performance (even more so than today).

  • Creating custom things in the digital realm is now within everybody's reach. Will development become a commodity? My guess is that the ability to create digital products will become even more of a power law.

• AI will transform biology before it transforms robotics.

• We can’t download knowledge into our brains just yet, but we’re bound to get access to an external, second brain based on our notes and online behavior

🤝 How we organize

• This article by Gregor Hohpe makes a compelling case that a software architect’s time is better spent debugging than coding.

• The permissionless corporation - HBR must be reading our newsletter.

• Here is a fascinating talk on spreadsheets: our most practical, most widespread interface to computation.

  • Related: I made a habit-tracking spreadsheet: feel free to make it your own.

In episode 6, I talked about social echo chambers in the context of our increasingly VUCA environment. At the time, did you wonder how one dominant opinion could emerge while others disappear? Or how opinions shift and cluster around certain ideas? Could we create a model to help us understand this system?

Physicist and complex systems researcher Dirk Brockmann has created an impressive collection of interactive models of complex systems in biology, physics, mathematics, social sciences, epidemiology, ecology, and other fields. One of them illustrates the phenomenon of opinion dynamics:

Unfortunately, I cannot embed the interactive model in this publication, but I encourage you to try it out for yourself on the explorables website. There are no sign-ups - not even a cookie banner!

The model begins with a set number of nodes representing people, and links connecting them. If two people with different opinions are connected, one of them might change their opinion to match the other person's. Alternatively, the link between them might be broken, and one of the people will connect with someone else chosen at random.

The sliders - enabled by computation - let you control the probability that the link will be broken (rewiring), and whether people prefer to connect with like-minded people or not (open-mindedness).

If you set the rewiring probability to high (i.e., you make it more likely to break the link) and openmindedness to low, the group will quickly split into smaller groups with uniform opinions. On the other hand, if you make it less likely to break the link, one opinion will eventually become dominant.

I’m sure some readers are starting to question my promise of writing more actionable content (or even my ability to do so), but there are lessons here that apply to anybody trying to manage a complex adaptive system like a company:

• First of all, opinion dynamics play out in companies as well. Companies that encourage a diversity of opinions will learn faster than monocultures. The presence of diverse opinions can help challenge assumptions and beliefs, leading to a deeper understanding of issues and a stronger foundation for decision-making.

• Computational models can help us develop an intuition for complex system dynamics. While the dynamics can get technical (e.g., diffusion, fractals & self-similarity, flocking, contagion, network growth…), they can result in very tangible conclusions:

  • This simulation on collective intelligence shows how a school of fish can collectively find an optimal location

  • This Autobahn simulation demonstrates how speed variability causes traffic jams.

Static models like causal loop diagrams helped us analyze a system, but a computational model allows us to test specific hypotheses or explore the behavior of a system under different conditions.

This is a crucial aspect of modeling. If we do not formulate hypotheses or predictions that we can test, how can we know if our model accurately represents reality?

Things get even more interesting when we start syncing the model with the real-world concept it represents. Take digital twins for example. All models mimic reality, but digital twins do so by receiving data from physical sensors in the actual system. They constantly mirror the real system’s state.

Formula One is an example of an industry where digital twins have changed the game. Gone are the days when speed and aerodynamic design were enough to set a team apart from the competition. In a sport that has become synonymous with continuous technological improvement, teams have turned to digital twin technology to get a leg up on their competition. Red Bull and Mercedes, the dominant teams of the past decade, have invested heavily in this tech.

The younger generation of drivers, like Max Verstappen, have honed their driving skills using racing simulators, which are a relatively simple form of digital twins. As an anecdotal aside, Max Verstappen won his first F1 race before he even had a driver's license. The digital revolution is changing every game.

The drivers are merely the public interface of a F1 racing team. Behind every driver pair are hundreds of very smart and very competitive people, who all rely on digital twin models to understand how the car behaves on the track. Cars are continuously collecting and sending data, allowing engineers in the team to model the car's performance as well as strategies. 

Digital twins allow the teams to test and validate their understanding of the racing system by comparing their predictions of the model to real-world data. If the predictions of the model match the data, the team gains confidence in their understanding of the system. If the predictions of the model do not match the data, they can use this information to refine and improve the model.

Outside Formula One, the digital twin approach is popular in industry, telecom, and manufacturing contexts. Done well, digital twins can optimize the performance of physical assets, and predict maintenance needs or the behavior of these assets under different conditions. 

One example closer to home is Skyline Communications, a Flemish company that has created a global product that acts as a digital twin for operations in the ICT media and broadband industry. They may fly well below the radar, but that doesn’t keep them from achieving significant, consistent growth and impressive financials.

While I’m clearly a fan of modeling and computation, there are limits to what it can achieve. I’ll touch on those next week before coming back up the stack.

Share

The best of the rest

• While I still write in the traditional way - through coffee, sweat, and tears - more and more people turn to machine learning models to automate the production of content.

  • Huggingface has some interesting chatGPT prompts you can use for all kinds of creative purposes (have ChatGPT act as a spreadsheet builder, a screenwriter, a debugger, a plagiarism checker, a cybersecurity specialist…)

  • One engineer created a script that summarizes Arxiv papers and generates podcasts from these papers.

  • All this raises some philosophical questions. When content becomes abundant, attention becomes scarce. How do we avoid drowning in content sludge? And what moats can companies still claim? Community? Meaning?

• Tech debt (or UX debt) is known to creep up on teams. John Cutler illustrates the process in the visual below. When stakeholders ask for a business case to refactor a product, consider also providing a business case for a total rewrite - it is just around the corner:

• Speaking of tech debt, Southwest Airlines suffered a holiday meltdown in a classic story of public companies prioritizing short-term stock prices over healthy software foundations. The longer companies ignore these signals, the higher the price they pay. In this case, it was a planning crisis. Next time, maybe ransomware and a total lockout. Ignore it long enough, and the system collapses.

Further reading

• “Complex Adaptive Systems: An Introduction to Computational Models of Social Life”: a very readable yet scientific introduction to computational modeling by John Miller and Scott Page

• Dirk Brockmann’s complexity explorables

• “How Nature Solves Problems Through Computation” - an article in Quanta Magazine featuring data scientist and evolutionary biologist Jessica Flack

One way to get there is to upgrade the causal loop diagram to a stock & flow diagram. In order to apply this method, we must add a fundamental distinction to the variables that we plot on our diagrams. We need to classify every variable as either a stock or a flow. Understanding the difference is key to understanding the dynamics of a system.

A bathtub analogy is the easiest way to distinguish between stocks and flows. A stock describes the state of the system at a given point in time: the amount of water that has gathered in the bathtub. We measure this amount in a unit like liters. Some other examples of stock variables are inventory, cash, and population.

A flow, on the other hand, describes the change over a period of time: the amount of water entering or leaving the system. In our bathtub example, we use the same measure of volume (liters), but we include the rate at which they enter or leave the system, e.g., liters per minute. 

Over time, stocks accumulate or deplete based on flows. This allows us to represent the rate of movement into and out of the system. Some examples of flow variables include cash flow (a flow affecting the stock variable of cash), layoffs (affecting employees), net income (affecting retained earnings), etc.

A stock and flow diagram is a more formal representation of the causal loop diagram (and a necessary step for any attempt at computing the system dynamics). In the example below, I have created a simple diagram that keeps track of a company’s headcount:

But does it beat a spreadsheet? In simple processes, perhaps not. When things get more complex, the stock and flow diagram does have significant advantages. For one, you can extend a stock and flow diagram without losing sight of the system. Most business processes are not stand-alone formulas. We end up with a better model by articulating the interactions and feedback loops.

There are myriad ways to add to the simple model above. You could double-click on the factors that impact the headcount stock, for example, by adding more detail to the attrition outflow (retirements, promotions, involuntary exits, regretted leavers,…). You could also zoom out and map the impact of employee headcount on financial stocks and flows.

Of course, spreadsheets can also be endlessly extended, but they don’t force their users to think about the interaction between stocks and flows. As the systems thinking episode made clear, there is more leverage in identifying patterns and trends than there is in managing components. Spreadsheet managers may miss:

• Delays: A stock and flow diagram can show the time lag between different flows in the system, such as the delay between hiring new employees and those employees becoming fully productive.

• Cyclical patterns: A stock and flow diagram can reveal cyclical patterns in the system over time, such as seasonal fluctuations in hiring or resignation rates.

• Threshold effects: A small change in one element can lead to a significant change in another element. For example, a sudden increase in leavers that post unfavorable Glassdoor reviews might lead to a sharp decrease in the number of applications, which could have a cascading effect on the overall headcount of the company.

In summary, by including the stock & flow dimension, we gain an understanding of how the interactions influence the system's behavior over time. By modeling the rate of change and any time delays between interactions, we can play around with different scenarios and reason about the evolution of the system over time.

Once a model becomes a program, we can bring the full force of information technology to bear on this representation of reality. That may seem trivial, but the ability to simulate how a system responds to different inputs and conditions is a superpower. I’ll zoom in on this in next week’s episode.

Further reading

This episode, as well as the last, builds heavily on two sources:

• “Systems Mapping: How to build and use causal models of systems”: a very practical and open access book written by complexity scientists Alexandra S. Penn and Pete Barbrook-Johnson

• “Business Dynamics: Systems Thinking and Modeling for a Complex World”: the classic book on system dynamics by John Sterman

What is collective intelligence? It is a form of universally distributed intelligence, constantly enhanced, coordinated in real time, and resulting in the effective mobilization of skills.

My initial premise is based on the notion of a universally distributed intelligence. No one knows everything, everyone knows something, all knowledge resides in humanity.

— Pierre Levy

Leveraging collective intelligence toward action for good has proven harder than leveraging the madness of crowds toward disinformation and chaos. But the same is true of the analog world: it is always harder to build than it is to destroy. Our species is still figuring out the coordinating mechanisms that unlock our ability for networked cooperation.

That is the mission statement of this publication: to explore the mechanisms that can help us level up our powers of coordination. In the next few episodes, I look for ways to better understand the complex, adaptive systems we live in. Levy was right that nobody knows everything, but everyone knows something. How can we tap into this collective intelligence?

The smartest person in the room is the room itself

— David Weinberger

There is a reason why post-its and workshops are the go-to tools for consultants. Creating maps of systems is a way to harness the collective knowledge, resources, and perspectives of those around us. If we can make this intelligence visible and accessible to more people, we can accelerate our ability to learn and grow together. But how do we make a map of a complex system?

One expert explainer of complex systems is Simon Wardley. He invented Wardley mapping: a technique for strategy mapping that has played a significant role in the digital transformation of the UK’s Government Digital Service. I’ll come back in more detail on Wardley mapping in a later episode, but I want to start this exploration with some fundamental truths that Simon Wardley discovered about mapping:

The entire video is a masterclass in explaining and well worth watching, but I want to highlight two takeaways:

It is easier to argue maps than to argue narratives

A map is a factual way of exploring and explaining a landscape. An alternative approach is to use storytelling to make sense of our surroundings. Storytelling may be a great sales tool, but when it comes to decision-making and alignment, it is problematic.

Narratives are personal things, tightly coupled with people’s belief systems. When narrative lines are drawn in the sand, discussions tend to get messy and political. A disagreement about the lines on a map is generally more constructive.

Graphs vs maps

A graph is a set of nodes connected by lines. These nodes usually represent the components of the system (e.g., users of a social network, or animals in an ecology), and the lines represent the relationships or interactions between those components (e.g., friendships between users, or who eats whom). Graphs are most often used to represent and analyze relationships between different objects or concepts.

Maps, on the other hand, are used to navigate landscape and movement. That means we need to make sure that space has meaning in a visual representation:

As usual, there are no silver bullets; one method isn’t better than the other. Both mechanisms can help us get a better understanding of the systems we live in. Graphs focus on the analysis of relations between components, while maps focus on the position and direction of components.

What about feedback? In the last episodes, I established feedback as a key characteristic of complex, adaptive systems, but neither graphs nor maps focus on the feedback loops between components.

One way to articulate our understanding of the interconnections in a system is through a causal loop diagram. It is possible to link key variables together into sentences that indicate the causal relationship between them. By stringing together several of these loops, we can arrive at a coherent description of the system.

I’ll illustrate this idea with a well-known example of a dynamic, interconnected, and complex problem: the Covid-19 pandemic. 

A causal loop diagram attempts to map the most important variables in a given system. In the pandemic case, that might be the number of infections and the chance of viral transmission. These variables impact each other, and together they create a reinforcing feedback process. Whenever we see growth or decline in a system, there is reinforcing feedback at play: a variable continually feeding back upon itself.

Reinforcing loops do not endlessly reinforce their own growth (or collapse). Eventually, they run out of steam. Balancing loops are the yang to the reinforcing yin; they counteract the change happening in a system. They seek goals and push towards equilibrium. Whenever we see goal-oriented behavior in a system, there is balancing feedback involved, as illustrated in the loop I added to the diagram:

It is possible to extend causal loop diagrams as far as our comprehension of the system allows. In the pandemic example, we could add different vaccination scenarios to the model. In that case, relevant variables might be the availability of vaccines, the willingness to get vaccinated, different compliance policies, etc. 

At the end of the day, the selection of variables and the framing of their relationships are subjective. In the pandemic example, I included government-issued restrictions as a way to dampen the undesired reinforcing feedback loop. You may disagree and offer an alternative control loop instead. By articulating the expected causal loops, policy discussions become more factual and less entangled.

Policy and strategy are hard because we are looking for causal threads in a haystack of interacting variables. Will lockdowns suppress the virus? Will agile transformation programs result in shareholder value? Causal loop diagrams will not answer these questions for you, but they are likely to improve the rigor of thinking and snuff out biases and assumptions. 

If these diagrams look unwieldy, it might help to think of them as functions that describe how the system behaves over time. Reinforcing loops, for example, translate into exponential curves:

When we add the balancing loop, we expect it to ‘flatten the curve’ and act as a brake. We can also call this goal-seeking behavior:  

Practitioners of systems thinking and system dynamics learn that similar patterns of behavior show up in a variety of different situations. The underlying structures are always different, but they often result in a handful of characteristic patterns:

Next week I will dive deeper into the systems theory rabbit hole, and I’ll explore the art of modeling complex systems.

Further reading

• Knowledge ecologist Christina Bowen

• Expert Wardley mapper Hired thought

• A kindred spirit I recently discovered: Luke Craven

Mathematician and polymath Norbert Wiener was one of the first scientists to theorize that all intelligent behavior results from feedback mechanisms. He was a central figure in the origin story of complexity theory.

Norbert Wiener started in the field of control theory but extended it beyond engineering and mathematics. Early in his career, Wiener formulated a mathematical description of Brownian motion – the random movement of microscopic particles suspended in a gas or liquid. This breakthrough would allow him to predict the trajectory of these particles probabilistically. 

Wiener spent World War II applying his theoretical work toward the war effort. More specifically, he worked on the problem of anti-aircraft artillery: how do you hit a target that is moving at high speed, high altitude, and that is actively trying to avoid being shot at? By WWII, planes were flying almost as fast as the payload of the AA guns targeting them, and gunners could no longer simply point the gun at the enemy. The plane would be long gone by the time its shells had arrived at that location. 

Wiener came up with a computation device that translated the radar information of the target’s distance, direction, and speed into a prediction of the enemy aircraft’s flight path. This information was then fed back into the system, allowing the artillery to adjust its aim and fire at the predicted location of the aircraft. This feedback loop allowed the system to continuously update its predictions and improve its accuracy over time. 

Wiener’s device started every computation with a new data set, unique to that pilot and plane. In computer science parlance, we say that the computation happens at runtime. Variables like the plane's capabilities, wind speeds, and the pilot’s temperament were all abstracted away into a time series of radar data. Wiener’s algorithms could then translate that data into a setpoint for the anti-aircraft system: a position of the anti-aircraft gun barrel that would blow the enemy out of the air:

The wide applicability of self-correcting systems was not lost on Wiener, who was more drawn to developing the underlying theory than furthering military applications. Wiener grew convinced that the study of complexity should not focus on traditional physics concepts, such as mass and energy, but rather on concepts like feedback and information processing. 

He called this new domain cybernetics, derived from the Greek word for steersman. Wiener summarized cybernetics as “the entire field of control and communication theory, whether in the machine or in the animal.” Controlling systems through feedback is now commonplace, and it’s hard for us to appreciate the reach of Wiener’s insights.

Wiener’s ideas quickly spread through the academic world. A group of scientists coalesced in the famous Macy conferences. These interdisciplinary meetings rallied scientists from domains as diverse as physics, psychology, biology, and computer science. Its participants were interested in finding common principles underlying complex systems and were generally fascinated by the analogies between the born and the made.

While cybernetics has not succeeded in formulating a unified information theory for feedback, its influence on science has been profound regardless. Thanks in large part to its interdisciplinary nature, cybernetics has managed to question many long-held scientific assumptions. 

In a deterministic view of science, relative and probabilistic statements held little value. Thanks to Wiener and the people who followed in his footsteps, we can reason about systems in terms of circular causality, intentionality, feedback, hierarchies, and decentralized control.

In engineering, cybernetics would pave the way for countless industrial applications, from simple brakes and control circuits to complex systems like robotics and spacecraft propulsion. But cybernetics had more to contribute. Wiener’s definition of the field was novel in 2 ways: it extended control theory to communication (and therefore to information theory), and it did away with the distinction between man-made and living systems.

Cybernetics opened the door for technology to become more biological and for biology to become more technological. The idea of a brain-machine interface was first envisioned by Wiener himself in the 1940s. Advances in machine learning, robotics, and materials science have since led to proofs-of-concept of augmented human abilities.

Medical applications like the control of prosthetic limbs show that the brain-machine interface is no longer hypothetical. Elon Musk, the entrepreneur that is as prolific as he is controversial, is the co-founder of Neuralink, a company that is developing a chip that could be implanted in a person's brain. Neuralink also develops a robot to make the chip insertion process easier:

This will undoubtedly remind the reader of dystopian Hollywood movies, but cybernetics is not just about adding silicon concepts to biology. Technology is also being remodeled based on life.

Deep learning - a subset of machine learning - is fashioned after the way the human brain works. At its core, it uses a type of algorithm called an artificial neural network. These networks are comprised of small units called neurons that are connected together in layers.

Unlike traditional algorithms, which are explicitly programmed to perform a specific task, a deep learning model is trained using a large dataset of examples. The training process adjusts the connections between the neurons, known as weights, to optimize the model's performance on the task.

The key insight in the field of deep learning is that by increasing the size of the dataset and the complexity of the neural network, it is possible to achieve remarkable results, such as Copilot, Midjourney, and ChatGPT.

Another, lesser-known lineage of cybernetics led to fields like biomimetics and biomimicry. These fields apply principles from cybernetics to create technologies and systems that are able to adapt and self-regulate in response to their environment, just like living organisms do. For example:

• self-healing concrete that can repair cracks and damage over time

• solar panels that mimic the way leaves capture sunlight

• designing enhanced materials based on naturally occurring properties like hydrophobicity or antibacterial properties

When it comes to management and organization, there have been a few attempts to create new models based on the principles of feedback, control, and biomimicry. Before jumping to that cybernetic branch, I want to explore another question that can help us understand the implications of feedback in organizational systems. Is there any way we can formally describe the feedback loops in a system? Can feedback help us create better models of complex systems?

Subscribe now

In our quest for organizational performance, we want to gain a better understanding of the interactions that take place in a complex, adaptive system. Ideally, we want to get better at influencing systems such as teams, companies, supply chains, and perhaps even entire markets. Before we jump into the deep end, let’s start by exploring how control works on a more fundamental level.

A good place to start is control theory, a branch of applied mathematics that builds on feedback. Since the geek level is already over 9000, I’ll stay away from the actual mathematics. Fortunately, we don’t need to do the math to appreciate that many systems - both man-made and biological - rely on feedback processes. Some examples from nature:

• In biology, feedback loops regulate many organic processes. For example, the hormone insulin plays a key role in regulating blood sugar levels in the body. When blood sugar levels rise, insulin is released, which helps to bring them back down. This is just one example of a self-correcting feedback loop that helps to maintain homeostasis in the body.

• Something similar happens on the ecological level. In the population of predators and prey, a decrease in predators will lead to an increase in prey. A boom in the prey population means more food for predators, which will cause their population to rise again. 

• Fruit ripening is the result of a reinforcing feedback loop. When one apple starts to ripen, it will produce ethylene, a gas that will cause nearby apples to ripen as well, producing more ethylene. As a result, all the apples in a tree will ripen virtually overnight. 

• Climate change results from interacting feedback processes in the atmospheric carbon cycle. To name just one: melting sea ice exposes more ocean, which in turn absorbs more heat and causes more ice to melt.

Clearly, the concept of feedback is an important building block of biological systems. Why is this concept so foundational? To answer this question, let’s look at the evolution of feedback and control in man-made systems.

Since antiquity, humans have dreamed about automatons or self-operating machines. Truly automated control proved elusive until the 18th century when James Watt adapted an existing device (a centrifugal governor) to control a steam engine. Before Watt’s invention, steam machines needed human operators to manage the load placed on the engine by opening and closing a steam valve.

Watt designed a mechanical control circuit for this valve. It is made up of two leaden flyballs at the end of a pendulum. As the engine turns faster, the hinged flyballs fly outwards. This is a control feedback loop: when the arms of the pendulum separate, a linkage causes the throttle on the steam engine to close. This way, Watt’s governor reliably and automatically controls how much steam can enter an engine’s cylinders based on its desired running speed. This ensures constant revolutions per minute without the need for a human to babysit the machine.

The importance of this invention can hardly be overstated. Watt’s governor was instrumental in taming steam power and kickstarting the Industrial Revolution. The steam engine itself would be relatively short-lived, but the concept of a control circuit to regulate feedback could be extended to speed, pressure, temperature, and other process variables. It proved foundational for all automation (including the combustion engine). 

Governors were more than a mechanical innovation; they were the proof of concept for man-made control over matter and energy:

The difference between a car and an exploding can of gasoline is that the car’s information—its design—tames the brute energy of the gas. The same amount of energy and matter are brought together in a car burning in a riot and one speeding laps in the Indy 500. In the latter case, a critical amount of information rules over the system, civilizing the dragon of fire. The full heat of fire is housetrained by small amounts of self-perception. 

— Kevin Kelly, Out of Control

In other words, the mechanical governor was a primitive brain of sorts, endowing machines with situational awareness. This innovation redefined our relationship with technology. In lockstep with our worldview, our inventions could now start to shift from a linear and deterministic approach to the fuzzy logic of interdependent, complex systems.

Kevin Kelly’s work has been a major influence on this publication. Two ideas of his were especially foundational:

• Technology is increasingly merging natural and artificial characteristics. The world of the born is not separate from the world of the made.

• To take advantage of natural principles in technology, we have to let go of our traditional view of control. Supervising complex systems is futile if one disregards the feedback loops that play out between interconnected components in the system.

Feedback is a powerful concept for designing and controlling systems because it takes the system output into account. Before dismissing feedback as a concept you are already familiar with, consider the simple process of pouring a cup of tea. Clockwork thinkers see a linear process: I make tea, and I pour it into a cup.

System thinkers see a circular process, involving a number of variables that contribute to the goals of regulating the temperature, flavor, and hot water level of the beverage. The simple act of pouring hot water is actually a regulating system made up of 5 variables:

• the desired level of hot water in the cup

• the current hot water level

• the gap between the two previous variables

• the position of the kettle

• the flow of hot water

I won’t bore you with similar diagrams describing the regulation of tea flavor, temperature, sweetness, etc. The point is that we shouldn’t be too quick to dismiss a process as linear. In complex systems, innumerable feedback processes are at play, and each feedback process is a loop of continuously operating cause-and-effect relationships.

This circular relationship - the output of a system being used as input for the same system - has important implications for the behavior that emerges. One of the key characteristics of feedback loops is that they can amplify or dampen the effects of the input on the output, depending on the nature of the feedback. Feedback lies at the root of all shared characteristics of complex systems: emergence, information processing, and nonlinearity.

In the next episode, I’ll take a look at the evolution of feedback through the lens of scientific history. After that, I’ll explore how feedback loops can help us model the systems we create.

Subscribe now

So we live in an increasingly VUCA world now. That doesn’t mean we have to abandon all hope of controlling these systems. It does mean we have to switch strategies. The next episodes lay the groundwork by exploring the hidden order of complexity. 

Summarizing complexity theory is a daunting undertaking. Still, we’ll need some background to understand how complexity and systems theory can be applied to companies. The field of systems thinking sits at this crossroads and is a good place to start.

“Wicked problems result from a mismatch with how things work and how we think or perceive they work.” 

— Derek Cabrera

At the end of the day, we want to understand an organization’s ability to deal with uncertainty and change. Why are some organizations better able to adapt than others? Arie De Geus, a pioneer of systems thinking and a global executive at Shell in the 1980s, said that growth in a changing environment depends on institutional learning, which is the process whereby executive teams update the shared mental models of their organization, their markets, and their competitors.

The discipline of systems thinking attempts to help people construct these mental models. Lesser leaders reduce complex systems to individual components and events. They do not fully grasp how the different components interact. Systems thinkers, on the other hand, look for what lies beneath the surface:

• Events: what happened

• Patterns: what has been happening (finding trends) 

• Structures: why has it been happening (finding relationships between trends and events)

• Mental models: how do we think about what is happening (finding blind spots, assumptions, and beliefs)

The key to systems thinking is to appreciate any complex system as an interconnected whole. The components of the whole - the companies in the supply chain, the organs in the body, or the employees in a company - are interconnected and influence the behavior and emergent properties of the system. 

These components are not just connected but also interdependent: each part of the system depends on some other part for its effect on the whole. In the example of the human body: your brains need your heart and lungs to function and vice versa.

In commerce, every finished product is the result of countless people and companies collaborating in a complex supply chain system. To appreciate just how complex, it is worth reading “I, Pencil”, a delightful story about the coordination required to manufacture the humble pencil.

For those of us with shorter attention spans, Milton Friedman makes the point in a few minutes:

When there is interconnectedness and emergence in play, it is dangerous to study parts in isolation. What looks like a heart problem may actually be caused by another organ, or it may cause symptoms elsewhere in the system. The genius factor of fictional characters such as Sherlock Holmes or Dr. Gregory House is little more than a knack for systems thinking.

Reductionism - one of the tenets of the clockwork worldview and of Taylorism - is the opposite of systems thinking. If you take a complex system apart, it loses its emergent properties. Draper Kauffman stated it colorfully:

“…dividing the cow in half does not give you two smaller cows. You may end up with a lot of hamburger, but the essential nature of “cow” — a living system capable, among other things, of turning grass into milk — then would be lost. This is what we mean when we say a system functions as a “whole”. Its behavior depends on its entire structure and not just on adding up the behavior of its different pieces.”

— Draper Kauffman

In some cases, it can make sense to examine components of a system in isolation but not if your goal is to understand the system. There is an ancient Indian tale that illustrates the importance of perspective, context, and seeing the whole. Here is a summary of the parable:

Six blind men were asked to describe an elephant. The first man, feeling the elephant’s side, said the elephant was like a wall. The second man, holding the tusk, said the elephant was like a spear. The third man, feeling the elephant’s trunk, said the elephant was like a snake. The fourth man, feeling the elephant’s leg, said the elephant was like a tree. The fifth man, feeling the elephant’s ear, said the elephant was like a fan. The sixth man, feeling the elephant’s tail, said the elephant was like a rope. All six men were correct, but each one only knew part of the truth.

— Indian parable summarized by GPT3

In more recent history, the MIT Sloan School of Management has developed the beer game: a role-playing simulation game to illustrate some principles of systems thinking. On the surface, the game lets students live through typical coordination problems of supply chains. If you dig deeper, you will find that the game does more than that. Players experience firsthand that their actions can have unintended consequences and that their behaviors are predicated on how they perceive the system.

The game emulates a simple production and distribution system for a line of beer where players take up the role of a retailer, a wholesaler, or a brewery. During the game, players experience firsthand how small variations in initial demand cause greater and greater fluctuations as they reverberate down the supply chain.

This is because people in the retailer role are likely to misinterpret an increase in demand as an indication of future demand, leading them to place higher orders in anticipation of this expected rise. This signal can get amplified when wholesalers jump to a similar conclusion, ultimately causing the brewers to ramp up production. When the anticipated additional sales don’t materialize, this leads to excess inventory across the board. This is known as the bullwhip effect:

The game never fails to frustrate players who reliably shift blame to other parts of the chain or “erratic” customer demand. A key lesson is that we are a part of these systems. Our limited understanding of the whole chain causes the overreaction that brings the system out of balance.

Various systemic solutions have been conceived (shorter supply chains, lower lead times, more collaboration, inventory visualization across the supply chain...), but the bullwhip effect has not been eradicated. We need only look at supply chain disruptions following the Covid pandemic to see it in action today.

People learn quickly if cause and effect are closely related in time and space. The child that gets burnt has no problem identifying the link between the stove and the pain in their hand. In complex systems, the distance between cause and effect is greater, making it harder to connect the dots between events. 

When the connection between cause and effect is not linear - and it rarely is in complex systems - leaders should be modest in the policies they install. Unfortunately, the managerial class often selects for overly confident people, and many policies fall victim to the law of unintended consequences.

Once you start examining the world through a systems thinking lens, it becomes clear that an overwhelming majority of policy is little more than an attempt to fix the unintended consequences of earlier policies. Here is an example of a government wrestling with unintended consequences (in Dutch).

If our understanding of the system is too narrow and flawed, we cannot grasp the implications of our actions. When we think about the effects of our actions, we tend to divorce the intended or foreseen effects from the unintended consequences, the so-called side effects. What we call side effects is not an aspect of reality; it just demonstrates that our understanding of the system is incomplete!

So this brings us to a new question. How can we get a better understanding of interactions in a complex system?

Further reading

Peter Senge, The Fifth Discipline

Hardliners of the clockwork worldview might argue that humans or human-made systems are different. We are not ants or trees. We may have some traits in common with complex systems, but we have become conscious of our ability to think. Certainly, this awareness of ourselves and our environment makes us capable of planning and control? 

Our mind is capable of great things, but human intelligence and consciousness do not reduce the complexity of emerging human behavior. Rather to the contrary. Interactions between simple components like ants can lead to complex behavior. When conscious agents interact, the result is not less complex but more so. In ants, the emergent behavior is a living bridge. In humans, it is technology, civilization, and social media.

What is it about technology that increases volatility, uncertainty, complexity, and ambiguity (the VUCA acronym that has become a coping mechanism in managerial circles)?

In episode 03, I argued that digital disruption - specifically the ability to update software over the air - led to the displacement of the age of mass production. But digital disruption can be felt far outside of commerce.

One of the main reasons why information technology has made our society more complex is because it increases the number of interactions between participants in society. Social media and mobile have made us more interconnected and shaped the analog world in ways nobody predicted. 

We have only recently started to recognize the effects of social media and smartphones on our biological makeup. Consider the notifications you receive on your phone. When we receive likes or texts, our brain releases dopamine into our neural system. Dopamine is the same neural transmitter that is released after eating or having sex. This molecule evolved to make us feel good when we meet a survival need. 

Likes and text messages are not essential for our survival, but they do take advantage of a hardwired need for social connection. In other words, we have collectively opted into a technology that lets software jack into the pleasure center of our brains. As it turns out, there are downsides to giving away root access to our behavioral operating system.

At the outset, social media was a seemingly innocuous innovation. When Facebook was created as a network to connect friends, nobody would have guessed that liking and poking friends could have geopolitical ramifications like the Cambridge Analytica scandal or the Russian interference in the US elections.

How did we go from Farmville to MAGA in one decade?

As a rule, the Internet disintermediates. Producers and consumers have found new ways to connect to each other, thereby creating existential problems for gatekeepers, brokers, and mediators. With fewer middlemen, there is less moderation and more variance. The same is happening in the media sector:

• The baby boom generation had movies and 30-minute sitcoms. The XYZ Internet generations have those, plus YouTube channels, Netflix binge-watching, 5-second Tiktoks, and more. 

• Before the Internet, there were only a few sources of truth. A handful of newspapers and television controlled the narrative. Today, distribution is decentralizing. Everyone and their dog has a megaphone through Twitter, Substack, Reddit, etc.

Tech is eating the media's lunch, in terms of eyeballs and in terms of advertising revenue:

The battle over control of the media is one of the most important stories of this decade. Human behavior is governed by narratives in the same way as ant behavior is directed by pheromones. For most of history, humans used stories to signal information to each other. In a connected world, it’s no longer the actual story that matters, but rather the story that people tell about the story. This is called a narrative, and it is in the fires of media that narratives are forged.

For an example from current events, look at the story that is unfolding on Twitter. The events matter, but how they are framed and folded into a narrative is even more important. There is a mainstream frame, and then there are more contrarian takes. You be the judge.

(For a deep dive into the importance of narratives, I recommend Ben Hunt’s Epsilon Theory.)

The disintermediation of narratives leads to an epistemic dilemma. How can society find consensus around what is true in an environment that defies moderation and control? We cannot make sense of this dilemma without understanding the emergent properties of the social media ecosystem, like echo chambers. People inside an echo chamber have become immune to any ideas or arguments that invalidate their belief system:

Look under the hood of search engines and social media platforms, and you’ll find machine-learning algorithms that are mostly black boxes. While they may have been designed with seemingly good intentions - engagement and personalization - the output of the system can result in unintended consequences.

For example, the same search prompt can generate different results for different people, depending on what the algorithm thinks the user wants to see. Seen in this light, the Internet is not a window into the world, but a mirror that feeds us back our own biases.

The resulting filter bubble adds to the echo chamber effect. As people get into the habit of reading sources that confirm their pre-existing beliefs, they also start to self-select into groups with like-minded people. It is tempting to dismiss echo chambers as distant corners of the Internet, populated by flat-earthers and 9/11 truthers, but all media consumers are affected by this trend.

This very newsletter is a tiny blip in a small cluster of system-thinking nerds, attempting to reach people in a Taylorist galaxy, far, far away. Feel free to do your part in the culture war by forwarding this newsletter to your friends in bureaucratic organizations:

Share

As if filter bubbles and echo chambers weren’t bad enough, media companies - not just tech companies - are competing for attention and have the incentive to optimize for engagement. Since people are more likely to share content that makes them angry or upset, this results in a bias for outrage and emotion, adding to the volatility of our social climate. In summary, the current media paradigm creates echo chambers while pitting their inhabitants against each other.

I’m afraid I have more bad news. There is serious research by serious people that suggests that civilizations can collapse when they are overwhelmed by the amount of information they create. This is called the Information scaling threshold:

The internet has massively increased the complexity of our information environment, but hasn’t yet produced the tools to make sense of it. Old forms of social sensemaking—institutions, universities, democracy, tradition—all seem to be DDOS’d by the new information environment.

…

The timing here seems unfortunate. We’re facing planetary challenges: climate change, global pandemics, mass extinctions, increasing geopolitical tension, financial crises, looming nuclear threats… What a time to be hitting the information scaling threshold!

But, then, maybe the information scaling threshold is why we’re experiencing these crises in the first place? As our problems get more complex, our ability to meaningfully coordinate breaks down.

— Gordon Brander

Exploring the tools to make sense of this complex environment is the raison d’être of this publication. The Internet genie is out of the bottle, and there’s no going back. Instead of reverting to reductionist, linear, and deterministic tactics, we need to find a way to leverage these complex systems to our advantage.

Next episode, I’ll explore one such sense-making tool: systems thinking.

Further reading:

Subconscious

Thinking together

While analyzing a global history databank spanning 10,000 years, Shin, et al found a disconcerting pattern. Civilizations scale until they are overwhelmed by the information environment they create. This is The Information Scaling Threshold. Throughout the Holocene, societies developed additional layers of administration and more information-rich instrum…

Read more

4 years ago · 38 likes · Gordon Brander

Emergence & adaptation

A colony of ants is physically very different from a forest, an immune system, or an economy, but they all share an important trait. Every one of these systems exhibits complex collective behavior that emerges from the individual, interconnected components (ants, trees, cells, market participants). Their interactions give rise to complex and unpredictable behavior on a higher level. Our immune systems attack cancer cells, forests allocate soil resources, and economies enable price discovery.

Emergence enables the system’s ability to adapt. It is what sets complex systems apart from merely complicated systems like clockwork. Being adaptive means these systems can change their behavior to improve their chances of success. Adaptation can happen quickly, as is the case with a flock of birds in an evasive maneuver, or it can take place over long periods of time, as with human evolution.

Anthropologist Richard Wrangham argues that the human ability to cook food was the key factor in our evolutionary success. The newly emerged collective behavior of cooking made it easier for the human body to digest food. This was not just a competitive advantage vis-à-vis animals, but it also set off an evolutionary feedback loop that freed up energy for our brains to grow. Humans had learned to outsource parts of their metabolism to technology.

Technological progress only accelerated from there on out. Humans have since created technology that enables new kinds of complex, adaptive systems. Consider the Internet. This network wasn’t designed with any emergent, adaptive traits in mind but evolved them nonetheless. Systems researchers identified coevolutionary relationships between the different parts of the system like web pages and hyperlinks… As a result, the way information propagates over hyperlinks is adaptive.

To illustrate this evolution, think of the early days of the world wide web when the Internet was a network of read-only static web pages. The first major adaptation was so-called “user-generated” content, increasing interaction between participants. The web is currently undergoing another paradigm shift that revolves around concepts like encryption, ownership, permission, and identity.

The physicist Murray Gell-Mann said of complex adaptive systems that “although they differ widely in their physical attributes, they resemble one another in the way they handle information.” Let’s explore how that happens. 

Signaling and information processing 

Complex adaptive systems interact through signaling and information processing between interdependent parts. We can understand the moving parts in any system from just 2 types of feedback loops: self-reinforcing or self-correcting feedback loops. If a change in a variable amplifies whatever is happening, we call it self-reinforcing or positive. 

In ant colonies, individual ants that have identified a food source deposit a pheromone trail on their way back to the nest. When other workers pick up on these pheromones, they start to follow the trail to the food source, leaving behind a larger pheromone trail, etc.

For a more technological example of a reinforcing feedback loop: the larger the user base of iPhones, the more attractive the iOS platform becomes for developers. The more 3rd party apps the platform has to offer, the more Apple can grow the install base,...

Self-correcting or negative feedback loops counteract change and seek equilibrium. In the ant example, overcrowding at the food source or depletion of the food leads to fewer ants returning to the nest with food, which leads to a weakened pheromone signal, which leads to fewer ants following the trail.

Even when we have an inkling of the feedback loops in play, we tend to overestimate our understanding of the system. A linear view of the world keeps us from understanding the downstream effects of feedback loops, multiple interconnections, nonlinearities, and time delays between interactions.

Yellowstone is an example of the unintended consequences of human intervention in a complex system. In 1916 a government agency launched a program to eradicate wolves in the national park. As wolves disappeared, the elk population could rise unchecked, which led to soil erosion and reduced growth of willow and aspen. In turn, fewer trees caused the beaver population to shrink. 

In ecological terms, this is called a trophic cascade. The ecosystem collapsed because people with a limited understanding of the interconnections removed a single component from the food chain. Yellowstone's extermination of the wolves serves as a cautionary tale about our limited understanding of complex systems. On a brighter note: the wolves have since been reintroduced.

Non-linearity

Complex systems are non-linear; seemingly small changes can have significant and often unintended consequences. Non-linearity also brought the American meteorologist and mathematician Edward Lorenz to complexity theory. 

“Determinism was equated with predictability before Lorenz. After Lorenz, we came to see that determinism might give you short-term predictability, but in the long run, things could be unpredictable. That’s what we associate with the word ‘chaos.’ ”

— Steven Strogatz

In 1961, Lorenz was using an early computer to simulate weather patterns. Lorenz’s weather model included a number of variables, representing things like temperature and wind speed. At one point, he wanted to re-check some conclusions and restarted the simulation halfway through to win time. To do this, he had to manually enter data from a printout corresponding to the conditions at that point in the simulation. 

To Lorenz’s surprise, the simulation predicted entirely different weather than the initial simulation. As it turned out, the paper print-out used rounded numbers, so the numbers from the earlier simulation differed by a tiny fraction.

The effect of such a slight difference would be negligible in simpler, non-dynamical systems. In a complex system, this becomes impossible to predict. Scientists call this sensitive dependence on initial conditions. Here is a beautiful animation of how initial conditions can spiral out of control.

In a poetic turn of history, it seems that the name of the butterfly effect is not based on the visual similarity between Lorenz’s attractor and a butterfly but on the metaphor of a butterfly flapping its wings in Beijing, thereby causing rain in New York. How wonderfully strange that the graphical representation of this effect would also resemble a butterfly!

The butterfly effect is often misrepresented to make a point about leverage. It is not about tiny actions that can have big effects. The point is that the impact of an event in a complex system is not repeatable; we can’t predict what will happen the next time the butterfly flaps its wings (if anything).

This does not mean we should abandon all planning. Later in the season, I will tackle the implications of non-linear systems on organizational planning. Next episode, I’ll explore how information technology fits into the complexity worldview.

Further reading

Melanie Mitchell, Complexity: A Guided Tour

Apparently, it’s not easy to recognize the mental models that govern us. David Foster Wallace makes the point beautifully in this brilliant commencement speech:

“There are two young fish swimming along who happen to meet an older fish. The older fish nods at them and says: ‘Morning boys, how’s the water?’ The two young fish swim on for a bit and then eventually one of them looks over at the other and asks: ‘What the hell is water?’” 

— David Foster Wallace.

The point of the story is that the most obvious, important realities are often the hardest to see. One way to become aware of the water in which we swim is to question our assumptions and the foundational principles that lie beneath. This approach is known as first principles thinking.

What assumptions exist in organizational governance?

• The best way to manage an organization is through top-down command and control.

• Control happens on the functional level: an IT manager oversees the IT people, a marketing manager oversees the marketing people, etc…

• Management and planning are separate from operations and execution.

In the first episodes, I proposed that these assumptions are themselves rooted in the clockwork worldview of Modernism. Building on Newtonian mechanics and the discoveries of the scientific revolution, humanity had become confident in its ability to understand, control, and predict how the world works.

A quantum wrench in the clockwork

Newton was not wrong. In fact, we still rely on his theory of gravity to predict the motion of celestial bodies. The problem with the clockwork approach is that Newton’s mechanistic contributions explain some of the dynamics of our universe, but not all of them.

At the beginning of the 20th century, scientists like Planck, Einstein, and Heisenberg shone a new light on the physical laws that govern how objects move. Newtonian mechanics helped explain the movement of macroscopic bodies, while the newly-discovered quantum mechanics described the behavior of microscopic bodies such as atoms and subatomic particles. 

Heisenberg’s uncertainty principle is an important theory behind quantum mechanics. It states that you cannot measure a particle's momentum and position at the same time. This principle disproves Laplace’s demon: the idea that a super-mind could predict the future if it knew the speed and position of all the particles in the universe.

With that, the central tenet of the modernist worldview had been invalidated. The idea that the state of the universe at one time determines the state at all other times, turned out to be wrong.

Although the modernist worldview had lost its scientific footing, this did not immediately discredit the other ideas that had been built on the foundations of reductionism, linearity, and determinism.

Take Taylor’s clockwork approach to corporate governance; managers have grown quite comfortable cultivating the illusion of certainty. On Microsoft Teams, no one is in a hurry to question any foundational beliefs.

It took us half a millennium to replace the religious worldview with the modernist approach - and religion is still here. Similarly, the quantum update will take time to propagate through human operating systems (although technology might accelerate things this time around).

The main thesis of Season one is that most managerial and organizational practices are deprecated; they are grounded in a worldview that comes with a misplaced sense of deterministic control. We need a new way to make sense of the world we live in. Before we circle back to organizational practices, I want to offer another model of the world, one based on complex, adaptive systems.

From clockwork to complex, adaptive systems

It turns out that an alternative to the clockwork worldview has been staring us in the face all along. Nature offers us plenty of examples: flocks of birds, the weather, our immune system, forests, the human body… All these systems exhibit behaviors that cannot be predicted from their components. Intuitively we know that the clockwork metaphor breaks down at this point. 

Of course, I need more than intuition if I am to frame complexity as a basis for a new worldview. What does science have to say about complex systems? For now, there is no single science of complexity nor even an established definition. The field is fairly young and has its roots in different disciplines and domains. 

Absent a definition; the next episodes will look at complexity from different angles. Reassuringly, complex adaptive systems play a role in many scientific disciplines, both hard and soft. With such an abstract subject, it’s easy to get bogged down in theory, but I’ll take care to punctuate the story with examples.

Perhaps the most striking illustration of a complex, adaptive system is a colony of ants. The colony as a whole is capable of incredibly sophisticated feats of engineering. Ant nests make smart use of materials to create networks of underground passages and even regulate their brooding quarters' temperature. Colonies construct living bridges for different evolutionary purposes (scaffolding, defense mechanisms, predatory behavior…). 

How ants accomplish this, is radically different from the clockwork approach to organizing and building. Consider how a living bridge compares to its mechanistic, man-made counterpart:

“Bridges and buildings are all designed to be indifferent to their environment, to withstand fluctuations, not to adapt to them. The best bridge is one that just stands there, whatever the weather.”

— Andrew Pickering, The Cybernetic Brain.

Emerging, organic bridges, on the other hand, interact with their environment and become a part of it. They are the result of self-organization instead of central planning. Individual ants are equipped with minimal cognitive resources, and there are no colony architects or project manager ants to oversee the proceedings.

Biologists have known and studied this complex group behavior for a long time. A more recent evolution is that scientists have started to look for similar patterns in other fields. That is the scope of the emerging field of complexity theory. A logical place for researchers to start was to look for common properties among complex systems. 

The modernist worldview was built on determinism, linearity, and reductionism. What are the foundational principles and common characteristics of complex systems? And do they constitute a better model of the world? Tune in next episode for more nerdiness.