AI Has Been Loose for 33 Months. Yale's Latest CPS Update Says the Labor Market Still Has Not Cracked.

Thirty-three months after ChatGPT was released to the public, the U.S. labor market still does not look broken by AI. That is the headline from Yale Budget Lab's January/February 2026 CPS update, and it is one of the most important data points you can hold in your head right now [Fact].

If your mental model still says "AI is silently dismantling the labor market," the evidence keeps refusing to cooperate. Here is exactly what the Yale team found, what it does and does not mean for your job, and the one signal that did move.

What the Numbers Actually Show

The update's verdict is direct. Occupational dissimilarity, industry dissimilarity, and exposure and usage metrics all remain flat, lie within historical ranges, or continue along the trends they were already exhibiting [Fact, Yale Budget Lab].

The team also looked specifically at where the disruption narrative is loudest — recent college graduates in the AI line of fire. The dissimilarity between older and more recent college graduates rarely deviates outside of the 30-33% range since January 2021 [Fact]. Whatever AI is doing to graduate hiring, it is not yet pushing the occupational mix of 20-to-24-year-olds away from the occupational mix of 25-to-34-year-olds in any way that breaks historical patterns.

The bluntest single sentence from the report is the one to memorize: "Currently, measures of exposure, automation, and augmentation show no sign of being related to changes in employment or unemployment" [Fact, Yale Budget Lab quote].

That is the third way the data could have looked if AI were causing a labor disruption. It is not how the data looks.

The One Signal That Moved

The report is not a victory lap. The most notable difference in the latest data is an uptick in the dissimilarity of occupational mix between older and younger college graduates, though this remains at the high end of the historical range [Fact].

That uptick is the kind of thing you watch carefully without overreacting to. It is consistent with three different stories: AI is just starting to bend graduate occupational sorting, post-pandemic remote work is still settling, or normal labor-market noise is wide enough to look like a signal. The Yale team's framing — "high end of historical range" — is the right one. Not yet a break. Worth tracking monthly.

For workers, the practical read is this: if you are a recent college graduate, your occupational sorting is moving slightly differently from your older peers, but not catastrophically. If you are not a recent graduate, none of the macro signals are flashing red.

The Anthropic Usage Data Tilts Toward Automation

Buried in the update is a more uncomfortable point about how AI is actually being used right now. In March 2026, Anthropic released new usage data corresponding to February 2026, and both samples indicate that observed usage is more likely to be associated with automation than augmentation [Fact, Yale Budget Lab].

This is the gap to sit with. The labor market data says AI has not caused observable disruption yet, but the usage data suggests employers and workers are pointing AI more at "do this for me" than "help me do this." In economic theory, that should eventually show up in employment, especially for the most exposed occupations. It just has not yet.

Two interpretations fit. Either the macro data is too coarse and slow to pick up displacement that is genuinely happening at the firm level — a measurement-lag story. Or the convex cost curve of AI accuracy (the partial-automation argument from MIT and IBM researchers we covered separately) means that even automation-tilted usage produces less worker displacement than the deployment volume suggests.

Both stories are testable over the next 12 to 24 months. Both demand that you actually watch the data instead of vibing on either side of the debate.

What This Changes for the Way You Plan Your Career

Three working rules drop out of the Yale data.

First, do not plan your career around the strong-disruption story unless and until the macro data confirms it. The data has had 33 months to crack and has not. If you are deciding whether to pivot careers, retrain, or sit tight, the base rate is that the macro labor market is functioning normally for now. Your individual industry might be the exception, but the burden of proof shifts the other way: you need a specific reason to believe your slice is different.

Second, watch the recent-graduate occupational dissimilarity number. It is the most sensitive frontier indicator, and it is the one moving. If you are early in your career or hire from that pool, this is your monthly signal.

Third, the gap between rising AI usage and flat employment numbers is the most interesting variable in labor economics right now. The Anthropic usage data tilting toward automation while CPS data stays flat is the kind of contradiction that gets resolved either with a delayed jolt (employment catches down to usage) or with a quiet revelation that automation just produces a lot less displacement than the discourse assumes.

For most workers, the second outcome looks more likely given current data. But the data needs to keep agreeing for that to remain the right read.

What to Watch in the Next CPS Update

Three specific markers are worth tracking when Yale's next update lands.

The first is whether the recent-graduate occupational dissimilarity falls back into the middle of the 30-33% band or pushes through the top. A break would matter. A return would close one of the few open questions.

The second is whether the Anthropic usage data continues to tilt toward automation or whether augmentation use grows. Augmentation-heavy usage is the partial-automation equilibrium in practice; automation-heavy usage is the displacement scenario in slow motion.

The third is whether occupations the exposure indices flag as most-AI-exposed start showing employment effects that diverge from less-exposed peers. So far they have not. If they start to, the Yale team will see it before anyone else, and it will be the moment to update the mental model.

The labor market has been very stubborn about refusing to break in obvious ways. That stubbornness is itself the most important piece of data we have. Use it.

Sources

• Evaluating the Impact of AI on the Labor Market: January/February CPS Update — Yale Budget Lab (2026-03-15)

AI Assistance Disclosure

This article was researched using AI-assisted analysis of the underlying Yale Budget Lab report and corroborating external coverage. Direct quotations from the source are clearly marked. Editorial framing, the gap analysis between usage and employment data, and career-planning implications reflect editorial judgment for a worker audience. The 30-33% dissimilarity figure, the 33-month timeline, and the "no sign of being related to changes in employment or unemployment" quote are from the cited Yale Budget Lab update.