the collaboration principle

In “Company” published in 1979, Beckett returns to that refrain again and again throughout his piece , as if starting over were not a failure but the only honest way forward. It simply is the recognition that certainty was never coming, and the decision to begin again anyway, from nothing, as many times as it takes.

Leaders in the intelligence age are going to need exactly that discipline.

This is not a piece for executives alone. It is a call to anyone who leads anything right now — a company, a ministry, a movement, a coalition — to see, think and act differently.

Not a playbook to adopt if the numbers work out, but a small number of imperatives leaders cannot trade away, whatever the pressure to do otherwise, because leading exactly as before, on the assumptions of before, is no longer available to anyone.

Part 1 of this series made the case for four conversations organisations need to hold as AI reshapes how they decide, learn, create and compete — a case made as much for cabinets and coalitions as for companies.

This piece is about what leaders owe those conversations. Intelligence itself is not short of new models this year. Leaders are the ones who need them. These are the five imperatives we believe every leader owes this moment, whoever they are.

The Hype and the Data

---

Start with the evidence, because it is more sobering than most boardrooms, cabinets and executive committees have admitted. Three years of confident proclamations — mass unemployment, imminent AGI, a productivity bonanza — collapse against what has actually happened. And the reason isn’t that the technology is weak. It’s that pilots stall inside organisations that never change to use it.

The same story repeats at scale. The overwhelming majority of enterprise pilots fail to deliver a measurable return, and most organisations experimenting with AI right now see no bottom-line impact from it at all. Only a tiny fraction call their rollout mature.

The pattern is not confined to the private sector. Governments show the identical shape, for the identical reason: bureaucracy, centralised decision-making, and a workforce without the skills to govern the tools it is being told to use. The clearest illustration ran for four months in Washington, where a much-publicised push to slash federal spending through AI-accelerated efficiency delivered a small fraction of what it promised — disruption mistaken for redesign.

Civil society shows the same shape again: near-universal adoption, almost no strategic use. Most nonprofits have picked up AI tools individually and reactively rather than as part of any plan, and their leaders overwhelmingly point to the absence of direction from the top, not a lack of tools, as the real constraint.

This is the gap in which the hype lives. A former chief information officer who spent years running technology inside major retailers has given its two failure modes their names. The first is wishing: the belief that AI is magic, that you can wave its wand at a hard problem and skip the work of solving it — sincere, and dangerous precisely because it is sincere. Its insidious cousin is washing: claiming to be doing more with AI than you actually are, usually because a board, an electorate, or a donor base wants to see progress a quarter can’t yet deliver.

Both are common enough to be the norm rather than the exception, and both produce the same outcome: real cuts justified by a capability that doesn’t yet exist, followed quietly by the rehiring of the people who were let go.

That is not efficiency. That is a leader discovering, after the fact, that the capability they claimed never existed. The cost isn’t only financial — a meaningful share of the workforce, especially its youngest members, now admits to actively working against its own employer’s AI rollout. Trust, once spent this way, does not come back on request, whether it belongs to a workforce, a public, or a membership base.

And the world is not slowing down to let those models catch up. Half a billion years ago, the Cambrian explosion was the brief geological moment when life suddenly diversified from a handful of simple forms into nearly every major body plan that exists today.

Software is living through its own version of that moment: coding agents collapsing the cost of building so fast that the population able to build is expanding by orders of magnitude, tools and architectures multiplying faster than any single leader can track. Robotics is entering the same phase behind it. There is little reason to think the pattern stops there.

What is coming is not a market shift to manage. It is a Cambrian explosion of everything — and no leader, in business, government or civil society, gets to sit this one out.

---

At the Level of Discernment

---

The most useful finding in the research on AI’s limits isn’t that it’s weak — it’s where. AI performs brilliantly on structured tasks with a clear, checkable answer, and underperforms, sometimes badly, on the ambiguous, multi-step problems that depend on judgment and context.

The reason traces back to how these systems actually work: they predict the next word, not the state of the world. They hold no persistent model of how things causally relate to each other, which is why they struggle with reasoning, counterfactuals, and anything that unfolds over time.

At the Level of Honesty

---

One of AI’s most prominent researchers put it bluntly in early 2026: the entire industry has been swept up in its own hype. The discipline this moment calls for is testing bold claims before repeating them, not after — especially the claims made by people whose financial and reputational interests are served by exaggerating what the technology can do.

The same standard has to apply after the decision, not only before it. Averill tells the story of an employee who gave a company two decades of work before being told, in a short meeting, that software could now do her job. It couldn’t, not really — she was the person who knew why the numbers looked the way they did, the context no tool could see. The leaders who quietly rehired for roles they had just eliminated confirm the pattern from the other side: the efficiency never existed. Only the decision to blame something other than themselves for cutting it did.

What leaders owe here is honesty over performance — the same discipline applied twice, whether the claim is a company’s earnings call, a government’s savings announcement, or a nonprofit’s annual report: tested before it is repeated, owned after it is acted on. A model does not get to carry blame, or credit, that was never really its to carry.

At the Level of Structure

---

Where AI has actually worked, the pattern is consistent: it required rebuilding, not addition. The clearest gains show up only where AI was built into a redesigned workflow, never where it was layered onto an old one — and the organisations getting the most out of it are consistently the ones willing to rethink how the work gets done, not just which tool touches it.

Averill’s own account of running technology at two major retailers bears this out from the inside: the projects that delivered took time, people and real budget to fit around how the business actually ran, while the pilots that failed wanted clean, connected data and consistent decisions from a company that, like most real ones, ran on a dozen systems that did not agree with each other.

Government and civil society tell the identical story from the other side of the ledger. Public AI strategies routinely fail to translate into practice because the underlying structures — procurement, data, workforce skills — go untouched, and nonprofit leaders report the identical stall: initiatives that never scale past isolated tasks because nobody set the direction.

The same root cause shows up wherever pilots fail: organisations avoid the friction of genuine integration — the retraining, the redesign, the hard conversations about who decides what — and reach for the announcement that skips it instead. Barely a fifth of firms were running AI in production at scale as of early 2026, against the overwhelming majority that had expected to get there twelve months before.

That gap is exactly the redesign work leaders keep deferring — and deferring it forever carries a price already being paid. Leaders, of companies, of ministries, of movements, owe their institutions redesign over bolt-on. Anything less just produces a faster bureaucracy, not a different one.

---

At the Level of Collaboration

---

Ashby’s Law, which opened Part 1 of this series, still governs: only variety can absorb variety. A frontier this jagged, shifting by function and by task, cannot be navigated by a single point of view, however well-informed. It requires structured, cross-functional sense-making — the kind that turns scattered individual experimentation into a collective, integrated capability. Most organisations are still doing only the former, largely alone.

We have watched plurality do this work outside the private sector too. Part 1 of this series described an international coalition that found, in two days of structured deliberation, what years of formal coordination had missed. The mechanism does not check what kind of institution is using it. Leaders owe plurality over uniformity. It is not a cultural nicety here. It is the only mechanism precise enough to track a frontier this uneven.

At the Level of Renewal

---

The refrain that opened this piece belongs here too. Amid a Cambrian explosion of everything, no leader — including us, and including whoever is currently defending a government’s approach or a movement’s — gets to hold the same answer for long. Leaders owe renewal over preservation: the discipline to keep starting anew rather than defend what worked last quarter.

Finally on, from nought anew, as many times as it takes.

Immanuel Kant’s categorical imperative doesn’t ask what a leader can get away with. It asks what is owed — to the decision, to the people affected by it — regardless of convenience, regardless of who is watching. What if this became a universal law? Kant’s answer does not bend to circumstance: it holds, or it was never a duty at all.

That is the standard these five imperatives are written to meet — not a strategy to adopt if the numbers work out, but obligations leaders cannot trade away, whether they answer to shareholders, to citizens, or to a cause.

The organisations who wait for certainty before starting will be led by people still running yesterday’s decision models on tomorrow’s problem. The Collaboration Principle exists to help leaders — in companies, in institutions, in movements — build these five imperatives into how they actually think, decide and act — not as a framework to admire, but as a duty to practise.

If you recognise this in your organisation, your institution, or your cause — the sense that the old models are running out faster than new ones are arriving — we would be glad to have a chat with you.

---