The clearest way to show an AI return this quarter is to cut headcount. It is also the slowest way to earn one. Gartner surveyed 350 executives at organizations with at least 1 billion dollars in revenue that had piloted or deployed AI agents or autonomous technology, and roughly 80 percent reported workforce reductions. The finding that should stop a board in its tracks: the reduction rates were nearly equal among firms reporting higher ROI and firms reporting only modest or negative returns. As Gartner's Helen Poitevin put it, workforce reductions may create budget room, but they do not create return.
Two things are true at once here. AI is producing real productivity gains, and layoffs are a poor way to capture them. Both can hold, and confusing the first for a license to do the second is the mistake spreading across the market right now.
The pressure to make that mistake is intense. A WRITER study run with Workplace Intelligence, surveying 2,400 employees and C-suite leaders, found 60 percent of companies plan to lay off employees who cannot or will not use AI, even as only 29 percent of the same leaders said they see significant ROI from generative AI. Read those two numbers together. A majority are prepared to cut people over AI, and a minority can point to a return from it. The cuts are already moving faster than the proof that AI drove the value they are meant to represent.
Do AI layoffs actually deliver a return?
On the evidence, not reliably. A layoff lowers cost immediately, which flatters next quarter's margin, and that is the entire appeal. But a lower cost base is not the same as value created by AI. If a firm that cut deeply and a firm that cut lightly report the same returns, the cut is not what produced the return. That is what Gartner's near-equal reduction rates across high and low performers are telling us.
The limit is worth stating plainly, because we hold ourselves to it. Gartner's data is a survey of large enterprises, not a controlled trial, so it establishes a strong correlation, not that layoffs cause worse outcomes. It is enough to reject the assumption that cutting equals capturing. It is not enough to claim the reverse as a law.
The hollowing-out risk nobody prices in
There is a cost to over-cutting that never appears in the layoff math. Reduce a team past the point AI can actually cover, and output falls, institutional knowledge walks out the door, and the people you most wanted to keep leave on their own terms. The saving is booked on the day of the cut. The damage shows up two quarters later, disguised as a demand problem or an execution problem, when it was a capacity problem all along. Attrition risk and the loss of high-potential talent are exactly the signals a workforce view should surface before a cut, which is why they sit in the Human Capital Optimization view rather than in a headcount spreadsheet.
What is people amplification, and why does it win?
People amplification is the pattern Gartner found in the organizations earning the highest returns. Define it plainly: instead of using AI to remove people, these firms invest in the skills, roles, and operating models that let people guide and scale AI systems, so each person produces more and better work. The gain comes from higher output per person, not from a smaller denominator.
The difference between the two strategies is direction, and it is easy to miss on the surface.
| Replacement approach | Amplification approach | |
|---|---|---|
| Core assumption | AI substitutes for labor | AI is leverage on labor |
| Where the return is booked | Up front, as a cost cut | Over time, as output per person rises |
| What it optimizes | Headcount down | Output up |
| Primary risk | Cutting past capacity | Slower to show on a margin line |
| What it needs to work | A budget decision | Measurement of what people actually do |
Both strategies can appear on a headcount chart as fewer people. Only one appears on an output chart as more work. That is why the decision cannot be made from headcount alone, and why the AI Impact signal family treats AI adoption as a first-class, measurable signal rather than a line item.
Why can't leaders tell which one is happening?
Because they measure the things that are easy to see, headcount and spend and license counts, and not the thing that actually separates amplification from replacement, which is what people do with the tools once they have them. Vendor adoption dashboards report seats and logins. Seats are not behavior. A team can be fully licensed and unchanged, or lightly licensed and transformed, and a seat count cannot tell the two apart.
A defensible answer needs three things a login report does not have. It needs behavioral signal, drawn from the tools where work already happens, so the effect cannot be talked up in a survey. It needs a comparison, the adopting team measured against a comparable team that did not adopt, matched on tenure, role, and tool stack. And it needs disclosed confidence, so a leader knows how much weight the estimate can bear. That method is controlled cohort analysis with confidence scoring, and it is set out in the Levos measurement methodology. We report the size of the effect and the confidence in it, and we do not overstate a certainty we do not have.
Adoption is a signal, not a checkbox
Most measurement stacks treat AI adoption as a yes or no, either the seats are deployed or they are not. That framing is why the amplification-versus-replacement question stays invisible. When adoption is measured as a continuous behavioral signal, a leader can see the difference between a team that logs in and a team whose work has genuinely changed shape, which is the only version of the number worth acting on.
#### From assumption to measured effect
The practical shift is from deciding on a belief to deciding on a measured effect. Establish where AI is measurably lifting output per person, then let the leaner-team case follow the teams where the amplification is real and large. Where the effect is small or unproven, a layoff removes capacity and hopes the survivors absorb it. This is the same discipline finance already applies to every other line, and it is the same standard we argued for in why the AI ROI numbers in circulation fail the finance test.
The strategic read
The companies that win the next two years of AI will not be the ones that cut the deepest. They will be the ones that could tell, with a number they trusted, where AI was amplifying their people and where it was not, and staffed accordingly. Levos is a Human Capital Operating System, the intelligence layer above the tools you already run, aggregating the behavioral signals that no single tool sees on its own. It sits alongside your existing stack, not in place of it, and its job is to make the amplification question answerable before the layoff question becomes irreversible.
Layoffs will always be the fastest return to show. Amplification is the only one worth building a company on. The difference between them is not a matter of opinion. It is a matter of measurement.
Frequently asked questions
Do AI-driven layoffs deliver a return on investment? Not reliably. Gartner found about 80 percent of AI-piloting firms cut staff, but reduction rates were nearly equal among high-ROI and low-ROI firms. Layoffs create budget room, not return. This is survey correlation, not proof of causation.
What is people amplification? Using AI to make each person produce more and better work, by investing in the skills and operating models that let people guide AI, rather than using AI to remove people. The return comes from output per person, not a smaller headcount.
Why can't most companies tell whether AI is amplifying or replacing their people? They measure headcount, spend, and seats, not what people actually do. Amplification shows up in behavior, which requires signal from the tools where work happens, compared against a non-adopting team.
How should a leader decide where AI justifies fewer roles? Measure first, cut second. Establish where AI measurably lifts output per person using a matched comparison, then let staffing follow the measured effect rather than the assumption.
Measure amplification before you cut
Levos is opening early access to a small cohort of mid-market leaders who want to know where AI is amplifying their people before they make headcount decisions they cannot reverse.
Design partner cohort today: 150 to 500 employees. Expanding to 500 to 2,000 in the second half of 2026.
