In a lot of executive meetings right now, the value of AI shows up as a spreadsheet. A line for reduced labor cost. A projection of faster output. A target for efficiency. The numbers look disciplined, and a spreadsheet manufactures the feeling of certainty long before operations has earned it.

So the question gets asked, usually by someone who isn't in the work: if AI makes us more productive, why are we still carrying the same headcount? Sometimes the honest answer is that staffing can come down. More often it's more complicated, because the savings are easy to measure and the costs are not. Salaries sit on a ledger. Coordination drag doesn't. Neither does the approval that now takes a week, or the best person on the team deciding they're done holding a weakened system together.

The mistake underneath the spreadsheet is that it treats generated output as finished work. The tool produces something that looks complete, and activity spikes. But someone still has to check whether it's right and whether it should exist at all. That labor doesn't vanish because a machine wrote the first draft. It moves.

You can see it in any function, but it shows up most starkly in engineering. A lead who used to mentor builders is now a full-time reviewer of generated code, tracing errors and catching the plausible-looking thing that would have broken in production. From a distance the team looks faster. Inside, the review queue is where the work went. The same story runs through every function where the tool now drafts and a person now has to sign off.

Cut headcount before you understand that, and you tend to remove the people who were absorbing the ambiguity and catching problems before they spread. That work doesn't leave with the role. It concentrates on whoever's left, until a few trusted people become the bottleneck and every hard approval routes through them. The spreadsheet still shows the savings. Operations starts to feel brittle.

The management load goes up, too, not down. When roles shift this fast and output multiplies, someone has to keep re-clarifying what matters and protect people from tools that never sleep. Weak management in that environment gets expensive quietly — good people burning out while still looking productive, until the results finally show it and the savings story is much harder to unwind.

None of this is an argument against AI. Used well, it takes out drudgery and lets capable people work at the top of their range. The gains are real. The question is what you do with them: cut, or reinvest some of that capacity in the judgment and systems work the new speed is about to demand.

So before you restructure around AI, here's a more useful place to look than "how many roles can we cut?"

Those questions build a different company than the headcount question does. The cheapest organization on paper can turn out to be the most expensive one to run — it pays the difference in delays, rework, churn, turnover, and trust it has to buy back later at a premium.

AI will make leaner companies. Lean is not the same as hollow, and the leaders who can tell the two apart will be running something that still works after the savings have been booked.