Large companies expect artificial intelligence to shrink their headcount by 0.8 percent in 2026. Smaller companies expect the opposite: modest AI-driven employment growth, concentrated in skilled technical roles. Same technology, same year, opposite sign โ and that gap is the most operationally useful finding in a survey of nearly 750 corporate executives conducted by economists at the Atlanta Fed, the Richmond Fed, and Duke University (NBER Working Paper 34984, 2026).
The authors put the implication plainly: the shuffling of jobs "is not predominantly within-firm but rather occurs across the economy" (Richmond Fed, 2026). AI job reallocation is happening between employers, not inside them. If you run operations at a 50โ500 FTE company, that sentence describes a labor supply event heading toward your requisitions, and almost every mid-market AI plan I have reviewed this year is built on the assumption that it describes a threat instead.
The two numbers that invert the standard 2026 headcount plan
The survey ran in two waves: 603 responses from The CFO Survey panel between November 11 and December 16, 2025, plus 145 responses from senior finance decision-makers at Financial Executives International and NASDAQ member firms, collected from mid-December into January 2026 (Richmond Fed, 2026).
The aggregate employment finding is close to zero. Firms reported a negligible AI effect on headcount in 2025, and the average expected effect on 2026 employment is also near zero. The action is entirely in the dispersion around that mean: large companies expect AI to reduce their employment by 0.8 percent this year, while smaller firms anticipate modest gains.
The composition data points the same direction with a longer horizon. On average, firms expect the proportion of routine clerical workers to decline by 0.76 percent in 2026 and by 2.19 percent by 2028, partly offset by increases in skilled technical workers. But the decline is not evenly distributed. Firms with higher AI investment โ and large companies specifically โ are significantly more likely to cut their routine share, while small companies are significantly more likely to expand technical employment.
Read those two findings together and the picture is not net job destruction. It is a transfer. One set of firms is releasing clerical capacity and buying software; another set is hiring the technical roles that the first set is not creating fast enough.
What AI job reallocation looks like from the receiving end
The distinction matters because it changes which department owns the response.
If AI reduced employment within firms, the correct response would be a workforce plan: attrition management, redeployment, severance modelling. That is the enterprise playbook, and it is the one most mid-market operations leaders are copying, because it is the one that gets written about.
If AI reallocates employment across firms, the correct response is a talent acquisition plan โ and specifically a plan aimed at a labor pool that is about to become available at a discount to its replacement cost. The workers moving are the ones the survey identifies precisely: administrative work, data entry, customer service, and other routine operational roles are what executives most often expect AI to replace, while marketing, accounting, finance, management, and analytical work are what they most often expect it to enhance.
That is not a population of people who cannot work. It is a population whose previous employer decided that the routine 60 percent of their job was cheaper in software โ and who retain the firm-specific process knowledge, customer context, and exception handling that the software cannot supply. At 50โ500 FTE, that combination is expensive to build and cheap to buy right now.
The layoff headlines are real. They are also not about you.
The counter-evidence deserves a fair hearing, because it is loud and it is accurate.
AI has been the leading stated reason for U.S. job cut announcements for five consecutive months. Employers cited it in 112,713 announced cuts through July 2026 โ about 24 percent of all cuts this year โ and 10,970 in July alone (Challenger, Gray & Christmas, 2026). Anyone reading those releases would reasonably conclude that AI is a headcount event.
Look at where the cuts land. Technology announced 149,023 cuts through July, 31 percent of the 2026 total and up 67 percent year over year. The same report notes that AI-related cutting has been limited outside the tech sector, and that total announced cuts through July were down 41 percent from 2025 while hiring plans rose 25 percent. Challenger's own summary of the year is the useful one: AI is shifting the labor market, not dismantling it.
So both datasets are describing the same phenomenon from opposite ends. Large, AI-forward, mostly technology-sector employers are announcing reductions. Smaller firms across the rest of the economy are expecting to add skilled technical staff. The headlines document the release side of the transfer. Almost nobody is documenting the receiving side, because nobody issues a press release about hiring four people.
The cost-cutting case for AI is the weakest one in the survey
There is a second finding in the paper that should change how you underwrite your own AI spend, and it is easy to miss underneath the employment numbers.
Executives were asked what motivated their AI investment, rating each factor from 0 to 4. Improving production efficiency and labor productivity ranked at the top. Reducing labor and non-labor costs ranked lower. Innovation โ developing new products โ and demand-side motives, such as reaching and serving customers better, also outranked cost reduction. The measured productivity gains follow the same logic: they are not primarily driven by capital deepening but reflect increases in revenue-based total factor productivity, closely associated with innovation and demand channels (NBER Working Paper 34984, 2026).
Then there is the honesty check. Companies reported a 1.8 percent productivity gain from AI in 2025. When the researchers computed the implied gain from the component parts โ AI-attributed revenue change relative to AI-driven employment change โ the implied figure was substantially smaller across every major industry, in both 2025 and 2026. The authors name it a productivity paradox and attribute the wedge to delayed revenue realization.
Two operational consequences follow. First, an AI business case built on headcount savings is being built on the motivation that the executives closest to the spend rank lowest and that the data supports least. Second, if you are benchmarking your own results against reported gains from peers, you are benchmarking against a number that the same dataset shows is inflated relative to what shows up in revenue.
Where this evidence is thin, stated plainly
This is expectations data from finance executives, not realized outcomes from payroll systems. It measures what CFOs believe will happen by 2026 and 2028, and CFO forecasts of their own headcount have a known optimism problem in both directions. The 0.8 percent large-firm reduction is an average of intentions, not a count of separations.
It is also a single survey wave with a supplemental sample drawn partly from FEI, NASDAQ, and Duke alumni networks, which skews toward firms with an above-average interest in the topic. And the "small firms will hire" finding is a statement about relative propensity, not a guarantee of absolute growth.
What the data can carry is the direction of the asymmetry and the fact that the asymmetry is large. What it cannot carry is a precise magnitude for your sector. Size the opportunity from the direction; do not put the 0.8 percent in a model.
The capture plan, and why most mid-market firms will miss it
There is a structural reason this opportunity does not convert automatically: the firms best positioned to receive the talent are the least prepared to absorb it.
Census Bureau data show U.S. business AI use hovering between 17 and 20 percent from December 2025 through May 2026, with adoption rising steadily by firm size โ the largest firms are the heaviest users (U.S. Census Bureau, 2026). Job-posting evidence points the same way: the share of firms with at least one AI-mentioning posting rose from roughly 2 percent in 2018 to nearly 6 percent by the end of 2025, heavily concentrated among the largest employers (Indeed Hiring Lab, 2026). Mid-market firms intend to hire technical people. Most have not yet defined the roles those people would fill.
Three moves make the intent executable this quarter.
1. Write the two job descriptions before the market moves. Not "AI specialist." The roles the survey implies are narrower: someone who owns process redesign around a specific workflow, and someone who owns the verification layer where AI output meets a customer or a regulator. Both are hireable from the clerical-to-technical population being released. Neither exists on most mid-market org charts.
2. Re-underwrite your AI business case on the efficiency and demand lines. If your board deck justifies AI spend with an FTE reduction, replace it with throughput per FTE, cycle time, and revenue-per-employee targets. That is where the survey locates the actual returns, and it is the version that survives the productivity paradox.
3. Build a named-target list, not a job board strategy. You know which large employers in your region announced AI-attributed reductions in the last two quarters. Those announcements name functions. The people leaving hold process knowledge in the exact operational areas you are trying to strengthen, and they are reachable directly.
None of the three requires new budget this quarter. All three require someone to decide that the mid-market is the destination in this transfer rather than a bystander to it.
The decision for this quarter
Your competitors with 10,000 employees are running a headcount reduction. You have been reading their playbook as if it were yours. It is not โ AI job reallocation moves work between firms, and on the evidence, it moves it toward companies your size.
The decision is narrow: this quarter, do you write the two role definitions and re-underwrite the business case, or do you spend another quarter defending headcount nobody is coming for?
The reallocation is already priced. The only open question is which side of it you staff for.
