The Science of Steering

Most of what we care about sits inside a system too large to model. A company. A market. A body. A climate. We will never understand it fully. That’s not the problem. The problem is how to act inside it without pretending we do.

Three intellectual traditions have taken that question seriously, and each answers it differently. None is complete on its own. But together they describe what it takes to steer something you can’t see the edges of.

Cybernetics started in the 1940s with a deceptively simple question: what do a guided missile, a thermostat, a nervous system, and a company all have in common? Norbert Wiener’s answer was the feedback loop. A system senses its state, compares it to a goal, acts, senses again. The loop, not the part, not the plan, is the unit of control.

It doesn’t matter whether the system is built from neurons, transistors, or people in an office. What matters is whether the loop closes. Sensing wrong? Goal unclear? Action disconnected from sensing? The system drifts.

Ross Ashby gave the field its sharpest law: only variety can absorb variety. He called it the Law of Requisite Variety. If the environment throws ten kinds of problems at you and your system has three kinds of responses, you fail seven times out of ten. Simple arithmetic. To survive, the controller must match the complexity of what it’s trying to control.

This is why simple organisations collapse in turbulent markets. Not because they’re badly run, but because the range of situations they can respond to is narrower than the range the world keeps producing. Requisite variety explains the ceiling a solo founder hits around headcount five. The business has outgrown the controller. More variety than one person can match.

The power of the cybernetic frame is that it names the minimum anatomy of control: sensors, a reference, an actuator, and a loop connecting them. That’s it. And most things that look broken in organisations turn out to be one of those four missing.

The limit is that cybernetics assumes the goal is given. The thermostat doesn’t pick the temperature. The missile doesn’t choose the target. The founder does. Which is why cybernetics on its own will build you a very efficient way to do the wrong thing.

Complexity science took shape in the 1980s at the Santa Fe Institute, and it started with a different question entirely: where does pattern come from? An ant colony has no foreman. A market has no planner. A flock of starlings has no choreographer. Yet each produces structure that looks designed.

The answer is emergence. Simple rules applied locally produce global order that no single agent intended. Stuart Kauffman formalised this with fitness landscapes: imagine every possible configuration of a system as a point on a terrain, its performance as the height. Agents climb. Peaks are stable configurations, valleys are failure modes, and the landscape itself is the thing you need to understand.

What makes the landscape useful is that it moves. Your competitors climb it too, and the terrain deforms under their footsteps. A strategy that sat at a peak yesterday can be in a valley tomorrow, without anything changing on your side. This is why static analysis fails in dynamic markets. The map is being redrawn under the traveller.

Complexity also gave us path dependence. A system’s current state only makes sense as the product of its history. You cannot copy Toyota by studying Toyota today. The thing worth copying is the forty-year sequence of decisions that produced it, and that sequence is gone. The lesson for strategy is direct: timing and order matter more than analysis. The landscape you’re climbing is different for the early mover than for the late one.

The best-known result from the field is the edge of chaos. Systems are most adaptive when poised between rigid order and full randomness. Too ordered and they can’t respond to change. Too random and they can’t hold structure long enough to act. Every viable organisation lives on that edge. Most of the daily work of running one is staying there.

The third tradition is the strangest. Also the most useful. It comes out of developmental psychology, systems theory, and the work of Gregory Bateson, and it says something the first two don’t: the system is not out there. It’s a frame you chose to draw around a part of the world.

Metasystemic thinking is the habit of noticing the frame. Not “how do I fix the problem?” but “what made this a problem?” Not optimising the team, but asking why the boundary runs between this team and the next. Not debating the strategy, but asking what level of the business the strategy is pretending to be about.

This sounds abstract. In practice, it’s the difference between founders who keep hitting the same wall and founders who stop hitting it. The wall is rarely in the tactic. It’s in the level of description. Pricing problems are usually positioning problems. Positioning problems are usually identity problems. And identity problems? Usually unresolved decisions about who the company is for.

Climbing one level is almost always the move. But it’s expensive, because the previous level was comfortable and the higher level is vague. Most teams would rather optimise a broken frame than redraw it. The frame feels like reality. Redrawing it feels like chaos.

This is why it matters. It names what thinking has to do to work with complexity at all. Cybernetics gives you the loop. Complexity gives you the terrain. Metasystemic thinking gives you the move you have to make when the loop keeps closing on the wrong thing and the terrain keeps refusing to match the map in your head.

That’s a metasystemic demand, and it’s the one most strategy frameworks quietly skip, because there’s no template for it. You have to notice the frame you’re inside. Then choose to step out.

Stafford Beer was a British operations researcher who spent thirty years on one question: what is the minimum structure an organisation needs to stay viable in a complex environment? He was trained in cybernetics, took complexity seriously, and had the metasystemic instinct, an insistence on asking what level of the organisation was actually being discussed.

What came out was the Viable System Model. Five functions that any viable system must perform, from a cell to a nation state. Beer numbered them.

System 1 is the operational units that do the work: the teams, the products, the frontline. System 2 coordinates between them so they don’t collide: shared calendars, shared data, shared rules of engagement. System 3 manages internal performance: what’s the current state, where are the bottlenecks, who needs more resource. System 4 looks outward and forward: what’s changing in the environment, what’s coming that we don’t have a response to yet. System 5 holds identity: who we are, what we value, what we refuse to do.

Most startups have System 1 and System 5. The founder holds identity; the team does the work. Systems 2, 3, and 4? Absent. The founder compensates for all three at once, which is why the job feels impossible from about ten people onward. You’re trying to run the nervous system of an entire company through a single human spinal cord.

Beer also designed something most management theory has no name for: the algedonic channel. A direct line from anywhere in System 1 straight to System 5, bypassing every intermediate layer. In the body, it’s the reflex arc. Touch something hot and your hand withdraws before your brain even knows. In an organisation, it’s the signal that has to reach the founder in minutes, not in the next weekly report. Most companies don’t have one. They have dashboards. The two are not the same.

What Beer understood is that viability is not a property of any single layer. Brilliant strategy on top of broken operations? Dead. Flawless operations without external scan? Overtaken. External scan without identity? Drift. All five functions have to be present, and each has to be handled at its own level, without the level above doing the work for it.

The three traditions each caught a piece of this. Cybernetics gave Beer the loop. Complexity gave him the reason loops fail: the environment generates more variety than any single controller can match, so variety must be attenuated at every layer and amplified at every response. And metasystemic thinking gave him the recursion: every System 1 unit is itself a viable system, with its own Systems 1 through 5 inside it. All the way down.

He left a motto. Absolutum obsoletum: if it works, it’s out of date. The organisations that endure keep rebuilding their own steering while they’re still moving. They never stop.