Someone asked me recently whether AI will reduce headcount.
My answer: not if you're running it right.
It's a reasonable question. AI tools are getting genuinely capable, and in an economic environment where every cost line is under scrutiny, the instinct to ask "how many people can we lose?" is understandable. But it's the wrong frame, and organisations that start there tend to end up somewhere they didn't intend.
The question that actually matters
The right question is: what can we now do that we couldn't before?
That's a different kind of question. It starts with capacity, not reduction. What new work becomes possible? What volume of work becomes manageable that previously would have required headcount you couldn't justify? What quality of output becomes consistent that previously depended on your best people having enough time?
At a company I work with, we're working through exactly this. The business manages a high volume of complex transactions across a regulated lifecycle: procurement, validation, tracking, retirement, reporting. The team that handles it is excellent. The question we're asking is not how to replace them but how to give them more throughput. The same people, handling a higher volume across the full lifecycle, without the kind of manual overhead that was previously unavoidable at scale.
That's what a multiplier looks like in practice. Not fewer people doing the same work. The same people doing more of it, better.
What happened when companies cut first
The organisations that moved fast on headcount reduction, assuming AI would cover the gap, mostly found out it wouldn't. Google made significant cuts. Others followed, citing AI capability as part of the rationale. Then several of them started hiring again.
The work didn't disappear. The judgement didn't disappear. The volume moved somewhere else, or it got dropped and someone noticed.
This is predictable. AI is capable of a lot, but it is not a drop-in replacement for a person. It can accelerate specific tasks. It can handle certain categories of work reliably. It struggles with novelty, with context that lives in someone's head rather than a document, with the kind of read-the-room judgement that experienced people develop over years.
When you cut the person without redesigning the work, you don't get the efficiency. You get the gap.
Where the multiplier effect actually shows up
The teams getting the most out of AI are not the ones that automated themselves to a skeleton crew. They're the ones that got clearer about what each person's time was actually for.
Take the mechanical work that eats into team time: reformatting, reconciling, reporting, first-draft generation. AI can absorb a significant portion of it. That's real. But the freed capacity has to go somewhere useful.
It can go towards the work that previously got squeezed out. The deeper customer conversations. The analysis that needed more time. The edge cases that were handled inconsistently because no one had bandwidth. The strategic work that kept getting pushed by the operational.
Or it can just get captured as cost reduction. Which is a choice, but it's a short-sighted one. You're trading compounding capability for a one-time saving.
The headcount question as a diagnostic
There's actually something useful in the headcount question, even if it's the wrong starting point. If your instinct when adopting AI is to reach for the org chart, it tells you something about how you think about people in relation to work.
If you see the team primarily as a cost to be minimised, AI looks like a mechanism for doing that. If you see the team as the thing that actually delivers outcomes, AI looks like a way to increase what they can deliver.
Neither view is unique to AI. It's the same tension that's existed in every technology wave. Process automation in the nineties. Offshoring in the two-thousands. Cloud and SaaS adoption a decade later. Each time, the organisations that came out ahead were not necessarily the ones who cut the most aggressively. They were the ones who redirected the saved capacity towards something that compounded.
The question worth asking instead
Before any AI adoption conversation turns to headcount, it's worth sitting with a different question: what is the team currently not doing that they should be?
Not what they're failing to do. Most good teams are working hard. What's getting deprioritised, deferred, handled inconsistently, or simply not attempted because there isn't the time or capacity?
That's where the multiplier has the most to offer. Not in reducing what exists, but in making possible what wasn't.
The goal is not a smaller team. It's a team that can do more of what matters, without having to grow at the same rate as the work.