The Accountable Animal

On risk, agency, accountability, and where we go from here.

A model can now do most of what used to define a professional career. Read the filings, run the numbers, draft the memo, and poke holes in its own conclusion faster than you can read them. The reasoning is finished. What isn’t finished is answering for it.

You can hand a decision to a system. You cannot depose a system, sue it, fire it, or sit it across a table from a regulator and make it explain what it was thinking. Reasoning turned into a commodity. Accountability didn’t, and it can’t, because accountability isn’t a skill that improves with scale. It’s a liability, and liabilities land on a person with a name, an address, and something to lose.

Every forecast about AI erasing white-collar work assumes the human was there to think. Mostly the human was there to be responsible. The analyst who gets fired when the model is confidently wrong was never paid for the arithmetic. He was paid to be the one you could fire. That function doesn’t vanish when the arithmetic gets automated. It gets heavier.

The relevant law here is Jevons, not Moore. When something essential gets cheap, we don’t pocket the savings, we consume far more of it. Cheap reasoning means orders of magnitude more decisions than we have ever made, including the ones that were never worth the deliberation before, when thinking cost more than the outcome was worth. Now they get made by the thousand, inside every firm, every day. That volume was impossible while judgment stayed scarce. Judgment isn’t scarce anymore. Every one of those decisions still resolves to someone who owns the outcome, and there is no Moore’s Law for owners.

This is what dismantles the org chart. Hierarchy was never mainly about communication. It was a rationing system for one scarce input: good judgment. Information climbed the pyramid and decisions came back down because senior thinking was the bottleneck and you couldn’t afford to spend it everywhere. Make judgment abundant and the rationing has nothing left to ration. What abundance does not touch is the other thing the pyramid encoded: who is on the hook when it breaks. Companies built to sort people by who knows the most will be rebuilt to sort them by who is willing to sign.

So the scarce, expensive thing is no longer the person who can figure it out. That person is becoming infrastructure, metered like bandwidth. The expensive thing is the person willing to put their name on an outcome a machine mostly produced and stand behind it when it fails. Underwriting, not analysis. Judgment was already the harder half of most jobs. Now it’s the whole job, and it carries the liability, which is precisely why it won’t commoditize.

None of this removes the human. It concentrates the human. One person with an AI workforce can stand behind ten thousand decisions a quarter that a department of forty used to make. The forty were doing the reasoning. The one is doing the thing the machines still can’t: standing there when someone asks who is responsible.

That was always the job under the job. The machines just stripped away everything it was hiding behind.