Key takeaway: No for most people already sitting in AI-exposed roles. Aggregate unemployment in those jobs has not spiked since ChatGPT. The clearer early signal is slower hiring into those same roles for young workers, so the ladder into the job is thinning even when the seat looks safe.
I keep hearing the same question in slightly different clothes.
Will AI take my job. Will the model eat the team. Will next quarter be the one where the org chart goes quiet.
The question assumes a desk that empties overnight. Most of the early data is pointing somewhere quieter, and more awkward: the desk stays filled, and the doorway gets narrower.
The seat that still looks occupied
If you already have a seat in an AI-exposed role, the apocalypse has been late.
A July 2026 SIEPR policy brief by Mahoney, McEntarfer and Wahal puts it bluntly: there is little evidence that AI is causing significant aggregate job losses right now. Unemployment among workers in the most AI-exposed occupations has risen, but not faster than among the least exposed. Since 2022, the most-exposed group is up about 0.77 percentage points. The least-exposed group is up about 0.85.¹
That is not a clean AI massacre. It looks more like a broadly softer labor market with AI painted on top of the fear.
Anthropic's Massenkoff and McCrory, writing in March 2026 with a new "observed exposure" measure that weights real automated use, find the same broad pattern on unemployment: no systematic rise for highly exposed workers since late 2022.²
A Chicago Fed working paper makes the related point from the other direction. Occupations with high AI applicability still saw employment and wage growth from 2019 to 2024. Exposure is not a mechanical death sentence for the role. Restructuring shows up before simple disappearance does.³
For people already inside many exposed jobs, the seat has held better than the headline.
I am not saying nothing is happening inside those seats. Tasks move. Tools change. The person who used to draft the first pass now edits a machine's first pass. We have argued elsewhere that the useful frame is often task decomposition and collaboration, not wholesale replacement, including at https://thinknarratives.com/articles/ai-in-2026-the-truth-about-ai-hype-job-loss-and-the-future-of-work. This piece is about a different pressure point: who never gets the first seat at all.
The canaries on the on-ramp
The sharper early signal is age and entry, not mass firings.
Brynjolfsson, Chandar and Chen, using ADP payroll data through mid-2026 in their "Canaries in the Coal Mine" work, report no evidence of widespread economy-wide displacement. Then they show the canary. Employment of young workers ages 22 to 25 in AI-exposed occupations sits about 19% below where it would be if it had kept pace with less-exposed peers. Experienced workers in those same occupations do not show a comparable gap. The channel is mainly reduced hiring, not a spike in separations. Where AI use substitutes for human tasks, the pain concentrates. Where it complements workers, employment is flatter or rising, especially for people with experience.⁴
Anthropic's hiring look lines up directionally. For ages 22 to 25, they find a roughly 14% drop in the job finding rate into more exposed occupations relative to 2022, barely statistically significant, with no comparable drop for workers older than 25. Their most exposed list includes computer programmers, customer service roles, and work heavy on analysis or data entry.²
SIEPR ties the same worry to the graduate pipeline. New-grad unemployment hit 5.6% in early 2026, up 1.6 percentage points from three years earlier. Junior work often packs the routine research, analysis and writing that tools now handle cheaply.¹
The ladder into the job is thinning even when the occupied seat still looks safe.
Call it the canary on the on-ramp. The bird is not the mid-career analyst who already knows the clients. It is the new hire who used to learn by doing the first draft, the first ticket triage, the first ugly spreadsheet.
SIEPR is honest about the messiness. Interest rate hikes, pandemic over-hiring and remote work all hit around the same window as ChatGPT. Some hiring weakness in exposed roles may predate the big models. After extra controls, Brynjolfsson and coauthors see the clearer entry-level story land later, into 2024, when adoption and capability were more real.¹ That caveat matters. A canary is a warning light, not a finished autopsy.
When the ladder story becomes a job story
I have to argue with my own headline here.
"AI will not take your job" is a decent summary of the aggregate unemployment numbers for people already employed in exposed work. It is a thinner comfort if you are 23 and trying to become that person. And it may not stay true if firms treat junior roles as optional forever.
Training is not free. Juniors absorb mistakes, ask basic questions and need review. If AI makes the first pass cheap, a manager under budget pressure can decide the training seat is the leftover, not the investment. That choice protects this quarter's headcount line. It also starves the pipeline that produces next decade's seniors.
A hiring freeze on the bottom rung can become a displacement story later, once the firm discovers it stopped growing its own experts.
There is a second complication on the layoff side. SIEPR notes healthy skepticism about companies that name-check AI while cutting. Some cuts look like cash freed for AI spend, or a cleanup after pandemic over-hiring, with AI as the polite excuse. Human resources voices in that synthesis talk more about role consolidation and hiring avoidance than about wiping whole occupations overnight.¹
So the ladder thesis can overfit too. Not every missing junior job is a robot. Not every "AI restructuring" press release is a robot either. The honest read is narrower: entry into exposed white-collar tracks has gotten harder in the data we have, while many existing seats have not emptied on schedule.
Productivity up, doorway closed
The productivity evidence makes the ladder squeeze feel less mysterious.
In the Brynjolfsson, Li and Raymond call-center study, a generative AI assistant raised overall productivity by about 15%, with gains concentrated among novices and less-skilled agents, who saw roughly 30% more issues resolved per hour. Highly skilled agents saw little gain, and response quality edged down slightly for them.⁵
That pattern is useful and slightly cruel. AI can make the new person look more productive on measurable tasks. A firm staring at that chart can conclude it needs fewer new people, or that it can stretch a smaller junior cohort further. Output per head rises. Headcount at the door does not have to.
The same logic shows up in coding and writing experiments summarized in the SIEPR brief: tools often help less experienced workers more on speed and quality for bounded tasks.¹ Help the novice, then hire fewer novices. That is not a paradox. It is a staffing spreadsheet with a new column.
I keep thinking about how we talk about tools and sleep elsewhere on this site, including at https://thinknarratives.com/articles/stop-asking-a-chatbot-how-you-slept: the interface gets louder than the underlying signal. Labor markets can do the same thing. The model demo is loud. The missing entry-level req is quiet.
And the cost of a stalled early career is not only a wage chart. It sits next to the broader squeeze on what a "good life" costs, the kind of pressure we traced at https://thinknarratives.com/articles/the-good-life-got-too-expensive. A thinner ladder is not only an AI story. It is a household story with worse timing.
What the ladder will look like in 2028
Firms will keep experienced seats in AI-exposed roles longer than they keep the training seats that used to feed those roles. They will not mass-fire every exposed occupation in the next two years the way apocalypse headlines keep promising. Young workers aiming at coding, support and writing-heavy tracks will face a thinner on-ramp, with more of the first-draft work absorbed by tools before a human ever gets the badge.
The question "Will AI take my job?" still deserves a plain answer for people already inside. For most of them, so far, no. The harder question is who gets to become them.
FAQ
Will AI take my job? For most people already in AI-exposed roles, the early evidence says no in the aggregate. Unemployment in the most exposed occupations has not risen faster than in the least exposed since 2022. The clearer pressure is on hiring into those roles, especially for young workers.
Is AI replacing entry level jobs? It is reducing hiring into some AI-exposed entry tracks more than it is emptying existing seats. ADP-based research finds employment for ages 22 to 25 in exposed occupations about 19% below a less-exposed peer path, mainly through fewer hires rather than more firings.
Why is unemployment not rising if AI is disrupting work? Because much of the early adjustment appears to be hiring avoidance and task restructuring, not mass separations. Workers who never get the first job do not always show up as unemployed in the same occupation series as people who lost a seat.
Which jobs are most exposed to AI right now? Measures differ, but repeated lists include computer programmers, customer service roles, and work heavy on analysis or data entry. Exposure is higher where tasks are automatable and already show up in real AI usage, not only in theory.
Are companies using AI as an excuse for layoffs? Sometimes. SIEPR notes skepticism from industry leaders and labor economists that many AI-branded cuts are really over-hiring cleanups or cash shifts into AI investment. Some narrow automation cuts are real. The label is not proof.
Does AI raise productivity while still hurting hiring? Yes, that combination is plausible. In one major call-center study, an AI assistant raised overall productivity about 15%, and about 30% for novices. Higher output per person can reduce the need to hire the next person.
Footnotes
¹ Neale Mahoney, Erika McEntarfer and Karsen Wahal, "What is really happening to jobs? Separating AI hype from reality," Stanford Institute for Economic Policy Research (SIEPR) policy brief, July 2026. Little evidence of significant aggregate AI job losses; unemployment for most AI-exposed workers +0.77 pp since 2022 vs +0.85 pp for least-exposed; new-grad unemployment 5.6% in early 2026 (+1.6 pp from three years earlier); notes skepticism on AI-washing layoffs; discusses interest rates, remote work and timing caveats around the young-worker hiring story; summarizes productivity evidence including call-center and coding/writing experiments.
² Maxim Massenkoff and Peter McCrory, "Labor market impacts of AI: A new measure and early evidence," Anthropic Research, 5 March 2026. Observed exposure measure combining theoretical LLM capability and real usage; no systematic unemployment rise for highly exposed workers since late 2022; suggestive ~14% drop in job finding rate for ages 22 to 25 into exposed occupations vs 2022 (barely significant); most exposed occupations include computer programmers, customer service representatives and data entry / analysis-heavy roles.
³ Federal Reserve Bank of Chicago working paper WP 2026-12 on AI applicability and labor market outcomes. High AI-applicability occupations saw employment and wage growth over 2019 to 2024; frames exposure as restructuring rather than automatic displacement.
⁴ Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," Stanford Digital Economy Lab, August 2026 update (ADP payroll sample through June 2026). No widespread economy-wide displacement; employment of ages 22 to 25 in AI-exposed occupations about 19% below the less-exposed peer path; channel mainly reduced hiring rather than separations; substitute vs complement AI use matters; authors describe results as early descriptive indicators, not closed causal estimates.
⁵ Erik Brynjolfsson, Danielle Li and Lindsey Raymond, "Generative AI at Work," Quarterly Journal of Economics 140(2), 2025, 889 to 942. Generative AI assistant for customer support agents: about 15% overall productivity gain, about 30% for novice and less-skilled agents; little gain for highly skilled agents, with a slight dip in response quality for that group.
Sources
- SIEPR policy brief (Mahoney, McEntarfer, Wahal, July 2026): https://siepr.stanford.edu/publications/policy-brief/what-really-happening-jobs-separating-ai-hype-reality
- Anthropic Research (Massenkoff and McCrory, March 2026): https://www.anthropic.com/research/labor-market-impacts
- Chicago Fed working paper WP 2026-12: https://www.chicagofed.org/-/media/publications/working-papers/2026/wp2026-12.pdf
- Stanford Digital Economy Lab, Canaries in the Coal Mine (August 2026 PDF): https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf
- Brynjolfsson, Li and Raymond, Generative AI at Work (QJE 2025): https://academic.oup.com/qje/article/140/2/889/7990658
- Thinknarratives, AI in 2026: https://thinknarratives.com/articles/ai-in-2026-the-truth-about-ai-hype-job-loss-and-the-future-of-work
- Thinknarratives, Stop Asking a Chatbot How You Slept: https://thinknarratives.com/articles/stop-asking-a-chatbot-how-you-slept
- Thinknarratives, The Good Life Got Too Expensive: https://thinknarratives.com/articles/the-good-life-got-too-expensive