AI Automation and the Labor Market 2025-2026
A survey
The U.S. unemployment rate remains low (4.3%1), but anxieties about AI-driven job loss run high2. Layoffs spiked in 2025, hitting the highest levels since the COVID-19 pandemic, but very little of this is attributable to AI automation. “AI washing” (the phenomenon of blaming AI for non-AI related job loss) is rarely-to-never done within corporate America itself. AI washing may occur in public economic discourse due to the high salience of advances in AI and the comparatively low salience of larger structural drivers of layoffs (e.g. restructuring, tariffs, DOGE, and — more speculatively — corrections to post-COVID overhiring). In the tech sector specifically, signs of structural unemployment driven by AI is growing. Financial analysts are generally optimistic about the next ten years of AI, grounded in the assumption that improvements in AI remain “normal”. However, significant improvements to agentic AI and the advent of recursive self-improvement would uncut the general assumptions of this optimistic picture.
What is the overall picture of layoffs in 2025-2026?
In January 2026, outplacement firm Challenger, Gray, & Christmas released its year-end layoff report3, which found the following:
1,206,374 job cut announcements overall in 2025.
54,836 (4.5%) job cuts attributed to AI by employers.
Public sector layoffs dominate. In particular, DOGE is the most common cause of layoffs. Market conditions, closing, restructuring, and cost-cutting follow.
In their January 2026 report4, layoff announcements in January are up 118% YoY. This spike is largely attributable to Amazon, which:
ended a relationship with UPS, causing 30,000 planned layoffs in the transportation sector.
slashed 16,000 corporate jobs5.
These latter layoffs are a part of a restructuring move designed to reduce “bureaucracy”.
These data suggest that the “AI washing” narrative is overstated, as employers are happy to attribute layoffs to non-AI causes.
Why be concerned about AI-driven job loss in 2026?
David Shapiro argues that the Challenger Report significantly undercounts AI job loss, and believes that the real figure is between 200,000 - 300,0006. He argues that employers are incentivized to underreport AI job loss (i.e. do “anti-AI-washing”), because failing to do so would “invite congressional hearings”. He derives his estimates from two methods, the first of which is the following:
Look at the delta in 2025 layoffs and mean annual layoffs from 2017-2019, and then chip away at it by removing chucks that are “cleanly explained” by things like DOGE, elevated interest rates, and tariffs. The remainder is 150,000-230,000 layoffs, which he attributes to AI.
This is not well reasoned. Layoffs that lack “clean explanation” should not be attributed to AI by default. Ambiguity cuts both ways. Surely one can conceive of other structural differences between the pre-pandemic American economy and the 2025 one.
There are a number of reasons to reject “AI-by-default” reasoning. The Yale Budget Lab found that an industry’s exposure to AI and their ChatGPT & Claude usage rates are not strongly correlated with changes in employment7. Consider for example the following graph:
Four curves are charted, corresponding to the length of unemployment. The y-axis measures the percent of AI task automation an unemployed worker was exposed to. If AI job displacement is a strong factor in layoffs, one should expect an upward trend, but this is not so.
Unfortunately, other plausible sources do find such correlations. Ozkan & Sullivan at the St. Louis Fed find that changes in unemployment rates in an industry correlate with AI exposure8:
It seems most likely that this discrepancy in results is attributable to methodological variation in how “AI exposure” is estimated.
In any case, Shapiro may be correct in his general point that AI job losses are undercounted (though, again, not to the degree he argues for). We can take the recent Amazon corporate layoffs as a case study: formally these layoffs are classified as “restructuring”, and indeed they are. But at least part of the thesis behind this restructuring is the idea that a flatter organization would be better situated to leverage ongoing improvements in AI. So, while one shouldn’t say that these tech restructuring layoffs are one-to-one AI substitutions, one should be open to the possibility that the ideal size of a technology corporation is already going structurally downwards.
Another reason to be concerned about AI job loss is that the post-COVID recent college graduate employment gap continues to widen, inverting historical norms.
Given the current state of the AI task exposure curve, one could easily imagine employers being both hesitant to deprecate an existing worker and hesitant to onboard a new worker.
“Normal” AI Progress vs. Slow Takeoff
If AI adoption rates continue to rise as they have for the last three years, I do not believe there is much reason to expect a big spike in AI-driven layoffs in the near-term. The labor markets of more and more industries will come to resemble how the tech industry currently looks — which isn’t too bad, all things considered.
However, one should expect adoption rates to speed up in the tech sector in the near term. METR shows that the time horizon of software tasks continues to explode:
We have seen that automating 30% of the tasks in an industry does not result in anything near the loss of 30% of the jobs, but I suspect this will categorically change once the percentage of automated tasks gets much higher. If a job is 30% automatable, it makes sense to keep the job and expand the role accordingly; if a job is 90% automatable, it makes much more sense to eliminate that job and to smear the remaining 10% across new roles that require fewer headcount.
The main metric worth tracking remains, for me, the percentage of AI R&D that is automated. Thomas Kwa at METR recently published a “medium timeline” model predicting 99% automation by ~20329. If we don’t live through this timeline, and we stay within “normal” AI progress, then we do not have much to fear with respect to AI-driven structural unemployment. But we are not on course for “normal” AI progress!
Bureau of Labor Statistics - “The Employment Situation - January 2026”.
Mercer’s Global Talent Trends 2026 report finds that “[e]mployee concern about job loss due to AI has surged from 28% in 2024 to 40% in 2026”.
Challenger, Gray, & Christmas - “2025 Year-End Challenger Report”.
Challenger, Gray, & Christmas - “Challenger Report: January Job Cuts Surge”.
CNN - “Amazon’s layoffs are staggering”.
David Shapiro - “AI destroyed 200k to 300k jobs in 2025 in the US”.
Yale Budget Lab - “Evaluating the Impact of AI on the Labor Market: Current State of Affairs”
annihilation 🤤