The Decision Budget

A follow-up to "More Done, More Tired," prompted by a chef who found the same problem in a completely different kitchen.

A Maître Cuisinier de France I worked with years ago recently told me how he analyzes a six-week menu cycle.

The old way: print every menu, tape them to a wall, and go at them with highlighters in a dozen colors, tagging repeated proteins, cooking methods, cuisine styles. Transfer all of it to an enormous spreadsheet. Get a team in a room. Walk the spreadsheet line by line until the critical points surface and an action plan falls out. Weeks of work. For a cycle that long, you might well have hired a consultant.

Now Pierre-Marie Leprince keeps a standing list of queries, pastes it in with the workbook, and has a complete analysis in about three minutes. Summary, areas of focus, recommendations. He also gets, as a throwaway, an assessment of how the cycle positions against the Mediterranean diet.

I wrote a piece a few weeks ago about being more productive than I have ever been and more tired than I have ever been. Pierre-Marie read it and replied with something better than what I had written, which is that the analysis was never the hard part. The funnel is. He now has a volume of information that nothing in the surrounding process was built to absorb, so by his own estimate he will act on a single-digit percentage of what is actually there.

That sentence has been rattling around in my head for a week, because it is exactly what is happening on my engineering team and I had not framed it that way.

The bottleneck moved and I did not notice

My last piece diagnosed the fatigue. AI removed the low-intensity work that used to be scattered through a day, the boilerplate and the waiting, and what is left is a concentrate of pure judgment. More decisions per hour than I have ever made. That is the tiredness, and I stand by the diagnosis.

What I got wrong was the prescription, mostly because I never wrote one. My unexamined assumption was that the answer is to build more capacity. Get better at absorbing. Read faster, decide faster, develop a stronger stomach for volume.

That is the wrong answer, and Pierre-Marie's single-digit percentage is why. If the constraint were capacity, then more capacity would help. But the gap between what these systems produce and what a human can meaningfully absorb is not a gap you close by trying harder. It is orders of magnitude. You cannot out-hustle it, and the attempt is precisely what burns people out while the useful fraction stays flat.

The bottleneck used to be producing insight. Now the bottleneck is metabolizing it. Those require completely different responses, and almost every team I talk to is still optimizing the first one.

A master chef and a CTO built the same gate

Here is the part that convinced me this is real rather than just resonant.

Pierre-Marie floated two ideas in response to his own problem. One was to box the number of decisions a person makes in a day or a given window. The other was to require proper discovery of what the AI has laid bare before anyone is permitted to make a decision about it.

That second idea is a comprehension gate. It is, structurally, the thing my team already runs in our development workflow: nobody advances a stage until they can answer questions proving they understood what the last stage produced. Not that it passed. That they understood it.

I built that to stop AI-assisted work from shipping unexamined. He arrived at it from a completely different direction, worrying about menu cycles and culinary R&D timelines collapsing from eighteen months to twelve to nine to six and now to something closer to instant. Neither of us was solving the other's problem. We landed on the same mechanism anyway.

When a master chef and a CTO independently invent the same control, the control is probably about something more fundamental than either domain. What it is about is this: the scarce resource is no longer information, and it is no longer output. It is attention that has actually been paid.

What a decision budget looks like in practice

Budget is the right word, and I mean it in the unglamorous accounting sense. A budget is not a target. It is a constraint you accept in advance so that individual choices do not have to be relitigated under pressure.

For us, in practice, that has meant a few things.

We put gates between generation and action, so nothing the AI produces converts into a decision without a person demonstrating they engaged with it. The gate is cheap. Skipping it is expensive and nobody notices for two months.

We batch decisions rather than interleaving them, because context-switching between unrelated judgment calls is where the fatigue compounds fastest. Half a day of one kind of thinking costs less than a day of six kinds.

We treat the weekly retro as the place where process decisions live, so they stop leaking into every hour of every day. Deciding when to decide is itself a way to spend less.

And we accept, deliberately, that we will use a fraction of what our tooling could tell us. This is the part that feels wrong to write down. There is analysis available to us that we will never look at, and options we will never evaluate, and that is now a permanent condition rather than a failure to keep up.

The part I have not solved

Rationing decisions protects the quality of the ones you make. It does not answer the harder question, which Pierre-Marie also raised and which I have been chewing on ever since.

He made the point that expertise is one of the critical things keeping humans meaningfully in the loop, and that it has to be protected, nurtured, and renewed. I agree completely. What I cannot yet answer is how the renewal happens. The grunt work that AI is now absorbing is, historically, exactly the work that turned junior people into experts. You learned judgment by doing a thousand small things badly and then less badly. If the thousand small things go away, where does the next generation of expert judgment come from?

I do not know. That is a whole piece on its own and I am still working out what I think.

What I am reasonably confident about is the smaller claim. The fatigue I wrote about last month is not a sign that you need to get tougher. It is a sign that the constraint moved, and that spending judgment as though it were unlimited is how you end up spending it badly. Decide less, on purpose, so the deciding is any good.

The volume is not going to stop. The only thing you actually control is how much of it you let become a decision.