Your AI spend per employee keeps climbing.

At what point does it cost less to keep the human in the loop?

CFOs are starting to run that math.

For the past two years, the pitch was simple:

AI replaces headcount. Headcount is expensive.

So AI saves money.

That logic is starting to wobble.

Not because AI does not create value.

Because the cost model is more complicated than the sales deck made it sound.

Here’s what’s actually happening:

1/ The swap is becoming explicit

→ CFOs are prioritizing AI and technology investment while headcount growth slows
→ AI budgets are increasingly being discussed as an alternative to future hiring
→ Leaders are asking whether to optimize employee count, AI spend per employee, or both

2/ The trade is moving faster than the proof

→ Layoffs attributed to AI are rising, though not every AI-labeled cut is truly caused by AI productivity
→ Many enterprise AI pilots still lack measurable P&L impact
→ Boards are pushing for efficiency before the business case is fully proven

3/ The per-head economics do not always favor the trade

→ For some heavy technical users, AI spend is starting to look less like software spend and more like a meaningful addition to labor cost
→ Nvidia’s Jensen Huang said he would be “deeply alarmed” if a $500K engineer did not consume at least $250K in tokens
→ MIT research suggests that broad replacement is not yet cost-effective across many categories of work

4/ Today’s “cheaper” may be partly subsidized

→ Some AI pricing appears strategically discounted to win market share
→ Flat-rate and promotional access can hide the true cost of heavy usage
→ A swap that works at today’s rates may look different when pricing discipline returns

5/ Usage will not hold still

→ A salary is relatively predictable
→ A consumption bill scales with usage
→ As agents take on longer, more complex tasks, token costs can rise quickly

6/ The people you keep may become more valuable, not less

→ You cannot cut the AI-fluent people you need to design, supervise, and improve the systems
→ Prompting, workflow design, model selection, governance, and QA become new operating skills
→ Some leaders are already talking about AI token budgets as part of the talent equation

7/ The savings only materialize if the work actually disappears

Cutting the role does not automatically cut the work.
Sometimes the work simply reappears as:
→ token spend
→ human review
→ workflow redesign
→ exception handling
→ governance
→ rework when automation fails

Workforce reductions free up budget.

They do not automatically create return.

The real question was never “human or AI.”

It is whether the work you are automating actually goes away, or simply changes shape and shows up on a different line of the P&L.

Swap a salary for a token budget and you may simply be exchanging a cost you understand for one you do not.

Save for future reference.