If you read the layoff headlines, AI is coming for the jobs. If you read the hiring data, AI is creating them. Both are true, and it’s worth sorting out before you plan around either headline.
US employers attributed 87,714 job cuts to AI in the first five months of 2026, more than in 2024 and 2025 combined, while the firms investing most in AI kept adding headcount. That’s the AI hiring boom: workforce recomposition, a changing job mix. The firm-level evidence is US-only; the sector evidence spans 27 countries.
Here’s the evidence, bottom up.
Why are the companies investing most heavily in AI also hiring the most, instead of cutting jobs?
For the heaviest adopters, AI spend behaves as a growth investment; the hiring effect sits at the top end of AI adoption intensity: depth of use, measured by per-employee AI spend, splitting adopters into heavy and light.
The layoff case is real. Block cut more than 4,000 roles and Oracle cut thousands, both citing AI, and tech layoffs in the past year passed 165,000 on the Layoffs.fyi tracker.
Ramp and Revelio Labs joined firm-level AI spend to workforce records for more than 21,000 US firms in A New Look at AI’s Impact on Jobs. Firms in the top third of per-employee AI spend, about US$30 per employee per month, grew headcount 10.2% over the two years after adoption. Low-intensity adopters showed no statistically significant change. Hiring also lags adoption by 6 to 12 months.
That dose-response is the clearest evidence in the study, and it comes with caveats: adopters were already larger, more engineering-intensive and more likely to be venture-backed, as Stanford’s policy brief lays out, and the sample is US-only and tech-skewed. Growing firms can afford AI spending and hiring, so you can’t read the 10.2% as a pure AI effect.
The 87,714 figure records what employers said, not what AI caused. Challenger keeps a separate “possibly AI” bucket, 20,219 cuts in 2025. More than half the firms blaming AI in layoffs were already growing AI headcount, per Revelio Labs. The label doesn’t always match the strategy; that’s the subject of how AI adoption is actually measured.
Two boundary conditions decide which way adoption tips: demand elasticity, whether cheaper output leads to more demand or fewer workers, and the super-star effect the PwC data exposes.
What did PwC’s 2026 Global AI Jobs Barometer find about AI exposure and headcount growth?
The most AI-exposed companies grew productivity 34% since 2018, against 24% at the least exposed, and grew headcount 52% versus 36%. But hiring splits in both directions inside exposed sectors.
PwC’s 2026 Global AI Jobs Barometer analysed more than a billion job ads across 27 countries and territories, which is how we know the pattern holds beyond US firm data.
AI exposure is a sector-level measure of task overlap with what AI can do, distinct from adoption intensity; it says nothing about whether a given company actually adopted AI.
Inside exposed sectors, the same exposure score maps to opposite outcomes: UK data since 2021 shows IT business analysts up 38% while call centre roles fell 19%.
The super-star effect: the top 20% of the most-exposed companies averaged 163% labour productivity growth, and 20% of companies capture 74% of all AI-driven value. Don’t build a headcount plan off the median.
One caveat: exposure correlating with productivity growth is not proof AI caused it. The full report and methodology notes are on PwC’s site. In Australia, AI hiring doubled between 2024 and 2025, and the most-exposed companies there are growing headcount at roughly double the rate of the least exposed.
So far, though, nothing explains why cheaper output creates more work. The mechanism comes next.
What is the “Jevons employment effect” and how does it apply to AI and knowledge work?
It’s the 1865 coal paradox applied to labour.
William Stanley Jevons observed in The Coal Question (1865) that more efficient steam engines did not save coal; they made coal power so cheap that Britain burned more of it, with output climbing from 50 million tonnes in 1850 to over 250 million by 1900.
Torsten Slok, Apollo’s chief economist, applied the same logic to knowledge work this year and named it. “When steam engines made coal more efficient, Britain didn’t burn less coal, it burned more,” he wrote in an April client note, pointing to cheaper legal, consulting and financial services. That’s the Jevons employment effect: when AI makes knowledge-work output cheaper and demand is elastic, output expands faster than labour input shrinks, so hiring rises.
Whether it applies to your market depends on demand elasticity. BCG’s split: software engineering is “amplified”, so cheaper builds mean organisations just build more and headcount holds or grows. Call centres are “substituted”: interaction volume is bounded by the customer base, so gains show up as fewer reps.
Two more mechanisms add work: new roles around deployment, forward-deployed engineers among them, and cheaper code expanding the problems worth solving.
Jevons only holds where demand can expand and displaced workers can reach the new work; where demand saturates, gains convert to cuts. Treat it as a hypothesis consistent with the 2026 data rather than settled economics. The deployment-side jobs are the subject of which roles are actually growing, and the mechanism’s limits are what make the checks below concrete.
What should CTOs check before assuming AI adoption will reduce their engineering headcount?
Five questions, in order.
Which tasks do you expect AI to absorb, not which roles? Tasks go before roles do.
Would the freed hours create new work? If your backlog, quality and product ambitions absorb the capacity, you’re in the elastic-demand world of the previous section.
What junior and senior mix does your delivery model actually need? US developer employment among 22-to-25-year-olds fell nearly 20% from its late-2022 peak while older cohorts grew, per Stanford’s “canaries” research; that’s the subject of the entry-level squeeze.
What do your own before-and-after numbers show? Same-engineer productivity, net engineering headcount and defect and review metrics beat vendor benchmarks; DX’s data across 400-plus companies found median pull request throughput up about 8%.
Are your peer benchmarks the right size and sector? The 10.2% figure carries the selection effect from the first section, so check any benchmark against your own size and sector.
Answer those five and you’re planning for workforce recomposition.
The two headlines resolve into one process: cuts and hiring happen at the same time because AI recomposes the workforce. The dose-response, the 27-country divergence and the Jevons mechanism all point the same way: plan for a changing mix. Explore the full series, and see how US and UK hiring patterns have split.
Frequently Asked Questions
Is AI actually causing layoffs, or are companies still hiring?
Both are true at once, but they describe different processes. The 87,714 US job cuts attributed to AI in the first five months of 2026 are real announcements, while firms investing most heavily in AI are net-hiring: high-intensity adopters grew headcount 10.2% over the two years after adoption. The reconciliation is Workforce Recomposition: cuts and hiring reshape the mix rather than uniformly shrink it.
My company’s rolling out AI: should I be worried about my engineering team’s headcount?
Worry about a blanket reduction, not about AI itself. Heavy AI adopters have generally kept hiring, but the effect depends on your demand elasticity and whether freed capacity creates new work. Before acting on fear, check which tasks AI absorbs, what junior and senior mix your delivery model needs, and what your own before-and-after data shows. Plan for mix change, not just headcount change.
Do only the heaviest AI spenders actually add headcount?
Yes, intensity is what predicts hiring. Firms in the top third of per-employee AI spend, roughly US$30 per employee per month in the first three months, grew headcount 10.2% over the two years after adoption, while low-intensity adopters showed no statistically significant change. That bar is modest, but it comes from Ramp’s tech-skewed client base, so treat it as a starting point for a 50 to 500 employee firm, not a target.
Does the record number of AI-attributed job cuts prove AI is destroying jobs?
No. The 87,714 US figure records what employers said, not what AI caused, and many of those cuts bundle AI with broader restructuring. A Stanford review finds little sign AI is causing significant job losses right now, and unemployment in the most AI-exposed occupations is not rising faster than elsewhere. Some firms are clearly AI-washing; Revelio Labs found more than half of the firms citing AI in layoffs had genuinely shifted towards AI roles beforehand.
Will AI replace software engineers, or will it mainly reshape the role?
The evidence points to reshaping so far. BCG classifies software engineering as an amplified role: AI accelerates coding, while system design, architecture, review and accountability stay human, and expandable demand keeps the work flowing. Ramp and Revelio Labs found AI adopters hiring engineers faster, not slower. The honest caveat is that frontier labs are targeting exactly this work.
Why do studies disagree about whether AI is helping or hurting entry-level hiring?
They measure different things and answer different questions. Ramp and Revelio Labs found entry-level headcount grew 12% at high-intensity adopters, while Stanford-linked research found AI dampening junior hiring in the most exposed occupations. PwC’s Barometer suggests a reconciliation: entry-level roles most exposed to AI are being “seniorised”, with those roles growing 35% since 2019 while other entry-level roles shrank 10%.
How do I measure whether AI is actually improving engineering output, not just activity?
Measure output and quality, not activity, and track three layers together: utilisation, impact and cost. DX’s research found median pull request throughput rose about 8%, with most organisations landing between 5% and 15%, even as tool adoption reached 93%. The signals that matter are same-engineer before-and-after productivity, net engineering headcount, defect rates, review queue depth and change confidence.
What are the warning signs of cutting engineering headcount too early?
Cutting before the learning curve plays out is the classic mistake: Ramp and Revelio Labs found hiring effects take 6 to 12 months to appear, so a reduction timed at month three is a bet on unproven gains. Other red flags include rising defect and review backlogs, shrinking junior pipelines, and cuts that remove judgement your AI-assisted workflow still depends on.
Does this evidence apply to Australian companies, or is it a US-only story?
It applies directionally, not literally. The firm-level dose-response evidence is US-only, drawn from a tech-skewed and venture-heavy sample, so the 10.2% figure is not an Australian benchmark. PwC’s sector evidence spans 27 countries and territories, which shows the direction is not uniquely American. Test it against your own before-and-after productivity and headcount data rather than treating it as a forecast.
Is the AI Hiring Boom guaranteed to continue?
No. The Jevons employment effect is a hypothesis consistent with the 2026 data, not settled economics, and it fails where demand saturates. Call centre and administrative roles are already contracting in several markets, and BCG estimates 10% to 15% of US jobs remain vulnerable to elimination over the next four to five years. The super-star effect is a further limit: the top fifth of firms capture outsized gains.
Where can I find the Ramp and Revelio Labs study on AI spending and headcount growth?
The study is Ramp Economics Lab’s working paper A New Look at AI’s Impact on Jobs, with Revelio Labs supplying the workforce data. It covers more than 21,000 US firms over two years, joins firm-level AI spend to workforce changes, and is explicit about its limits: US-only, tech-skewed and not a causal identification.
Where can I find the full PwC 2026 Global AI Jobs Barometer report and its methodology?
The full report, methodology notes and key takeaways are published on PwC’s website as the 2026 Global AI Jobs Barometer. The methodology explains what is measured from more than one billion job ads across 27 countries and territories, and the caveats that come with treating exposure and productivity as correlations rather than proof of causation.
