AI doesn’t replace the work. It moves the bottleneck.
Most AI commentary asks the wrong question. It asks what can the machine do now? — and the answer keeps growing, which is alarming if you’re keeping score that way. A more useful question is economic: when a capability becomes cheap and abundant, what becomes scarce and valuable? That’s the shift this essay is about, and it reframes the whole anxiety.
Generative systems are making content and execution close to free. Drafting, summarizing, coding a first version, producing a hundred variations of an idea — the cost of production is collapsing. When production gets that cheap, the bottleneck moves somewhere else. It moves to judgment: the capacity to look at a thousand machine outputs and decide which one is actually feasible, sound, on-brand, and aligned with what your organization is trying to be. The scarce resource is no longer the ability to make things. It’s validated judgment — discernment you can trust.
This flips the strategic question. For a century the operative question was “how do we produce more, faster, cheaper?” When production is nearly free, that question answers itself and stops being interesting. The new question is: how do we build the people and systems that can filter, test, and direct the torrent? Output is no longer the constraint. The constraint is knowing which output is worth anything.
This maps cleanly onto where, in any organization, humans remain essential. AI is genuinely good at stripping routine work out of the operational layers — the doing, the coordinating, the running of the present. What it doesn’t supply is the interpretation of ambiguous signals (deciding what a confusing situation actually means) and, above all, the identity-aligned discernment of whether something fits who the organization is. “Is this on-brand? Is this the kind of thing we do? Does this match what we stand for?” — that’s not a content question, it’s an identity question, and it’s the last thing you’d want to automate. As AI absorbs the routine, the irreducibly human work concentrates at exactly these two points: interpreting what’s happening, and judging what fits.
Which sets up a genuine fork in the road, and it’s a choice, not a forecast. One path is automation: use AI to engineer human judgment out of the system wherever possible, chasing efficiency. It works, narrowly — and it produces a hollow organization, one that’s fast and cheap and has quietly lost the capacity to discern, so it executes confidently in the wrong direction. The other path is augmentation: treat AI as a powerful new instrument that amplifies human judgment rather than replacing it. The organization becomes a kind of human–AI partnership — an assemblage — with its own emergent intent. And the decisive question about that partnership is whose intent is steering it. That’s not a technical question. It’s a question about identity and purpose — which means it belongs to the part of the organization that holds those things.
The honest caution: the line between machine-assistable and must-stay-human judgment is moving. Betting that identity-setting and final discernment stay human is, today, a safe bet. Whether it’s permanent is genuinely unknown — and it’s one of the live debates in this whole field rather than a settled conclusion. Anyone who tells you the boundary is fixed, in either direction, is guessing. The useful posture is to track the boundary deliberately as it shifts, not to assume it.
The takeaway is practical and a little contrarian: in an age of cheap everything, stop competing on output and start investing in discernment. The organizations that win the AI transition won’t be the ones that produce the most. They’ll be the ones that can tell, fast and reliably, which of the thousand cheap options is the right one — and that capacity is built in people and identity, not bought off the shelf.
Next week: even the best judgment runs into the oldest obstacle — organizational politics.
Part of the countdown to Metaphorum 2026 — “100 Years of Stafford Beer” (17–19 September 2026, Alliance Manchester Business School, Manchester), where I’m convening a half-day “Cybernetic Futures” workshop [workshop slot — TBC]. Programme, themes, and registration: conference2026.metaphorum.org. Abstract deadline: 17 July 2026.