Is AI Actually Taking Jobs? What August 2026's Data Shows
A Fortune investigation this week rounded up the studies. The layoffs are real, the picture is mixed, and almost none of the affected roles look like the jobs tracked on this site.
Every few weeks a headline claims AI just wiped out a wave of jobs. A few weeks later, a different headline claims the opposite: that AI is quietly boosting hiring. Both can point to real numbers.
That's the actual problem: the labor-market data tracking AI's effect on jobs is running months behind the technology itself, and right now nobody has a single number that settles the question. A Fortune investigation published August 8, 2026, by reporter Sebastian Herrera, pulled together the conflicting studies. What emerges is a labor market being reshaped in patches: some roles growing, some cut, and government data too slow to say which effect dominates.
What the Numbers Actually Say
The clearest data point comes from Ramp, the corporate-spend company, which tracked more than 21,000 U.S. firms over two years.
Heavy AI use correlates with hiring, not firing. Most companies aren't heavy users yet. A separate 2025 analysis co-authored by Stanford economist Erik Brynjolfsson found a real downside for one specific group: workers aged 22 to 25 in the most AI-exposed occupations saw employment drop 16% relative to their less-exposed peers. That's a young-worker problem more than a universal one, the kind of entry-level knowledge-work role AI can already do a passable job of, before someone has built up the judgment that makes an experienced worker hard to replace.
California's Policy Lab looked for a statewide spike in unemployment claims among AI-exposed roles since ChatGPT launched in late 2022 and didn't find one, except among college-educated workers in the most-exposed jobs, concentrated around San Francisco. Google's own research, for what it's worth, found AI functioning mostly as a collaborative tool rather than a replacement.
“It's been notoriously hard to pin that down.”
The Layoffs Are Real. So Is “AI Washing.”
None of this means the layoffs aren't real. Microsoft cut nearly 5,000 jobs in July while pouring billions into AI data centers. Amazon and Oracle have each cut thousands over the past two years. July's jobs report showed an unexpected 23,000 cuts.
The problem is figuring out how much of that is actually AI, and how much is companies correcting for pandemic-era overhiring and using AI as a convenient explanation.
“Companies definitely overhired during the pandemic and are now correcting, sometimes blaming it on AI.”
There's a name for the inverse of that problem too: a company avoiding any mention of AI in a layoff announcement to dodge public backlash, even when AI is the real reason. Nearly 200 economists, including Anthropic co-founder Jack Clark, signed a statement this year warning that AI could cause large-scale displacement within a decade without policy changes.
Where This Actually Lands, and Where It Doesn't
Almost everything in this debate (the Ramp data, the Stanford study, the California claims data) is measuring office and knowledge work: roles where a chatbot can plausibly draft the deliverable. It says close to nothing about a plumber tracing a leak through three walls, or a nurse deciding whether to escalate a concern to a doctor. That's not an assumption on our part. It's the actual shape of the exposure data.
Two jobs on this site carry real AI-usage numbers instead of a modeled estimate: elementary teacher and CNC machinist. One is a role AI genuinely touches day to day. The other barely registers. You can see exactly why on each job's page, and how we measure it, on our methodology page, rather than take our word for it.
The practical takeaway
If your work requires being physically present, adapting to conditions nobody wrote a manual for, or earning trust from people who let you into their homes or bodies, this August's numbers don't touch you much yet. If your job is mostly structured knowledge work performed at a screen (the entry-level version of it especially), the Brynjolfsson finding is worth taking seriously.
Either way, the honest answer this month is the same one we give on every job page here: read the structural exposure and the real-world usage as two separate signals, and don't let one contested headline stand in for both. Trades like electrician and HVAC technician, and hands-on healthcare roles like physical therapist, remain the clearest examples of work this month's data doesn't reach.
For a direct side-by-side on training time and pay, see our electrician vs. HVAC technician and physical therapist vs. registered nurse comparisons.
See Both Signals for Your Own Job
Structural exposure and real-world AI usage, reported separately, for every job in our directory, not a single contested percentage borrowed from a study measuring a different kind of work.
Frequently Asked Questions
Is AI actually causing layoffs right now, in 2026?
Some of it, yes, but economists say it is genuinely hard to isolate from a separate trend: companies correcting for pandemic-era overhiring. Ramp economist Ara Kharazian told Fortune that many firms "definitely overhired during the pandemic" and are now cutting back, sometimes attributing the cuts to AI when the real driver is the earlier overhiring. Real AI-driven cuts and "AI-washed" cuts are currently mixed together in the same headline numbers.
Which workers are actually affected by AI-related job loss?
The clearest documented effect is on young, college-educated workers in AI-exposed knowledge roles. A 2025 analysis co-authored by Stanford economist Erik Brynjolfsson found employment for workers aged 22-25 in the most AI-exposed occupations dropped 16% relative to their less-exposed peers. Broader, statewide effects have been much harder to detect in unemployment claims data.
How is StableJob's scoring different from these studies?
These studies measure employment and hiring outcomes after the fact. StableJob publishes two separate leading indicators instead: a structural exposure score (how much of a job's tasks a model could plausibly perform) and, where real data exists, a measured AI-usage band. Neither claims to predict who loses a job, which is exactly the gap this month's data confirms nobody has reliably closed yet.
Jobs Mentioned in This Post
View all jobs →How We Score Jobs
Two separate signals, reported honestly: structural exposure to automation, and real-world AI usage where the data exists, never averaged into one number.
Read our methodology →