Is AI really taking entry-level jobs in 2026?

Stanford's 19% AI Employment Gap: What Operators Should Do

Stanford's August 2026 revision of Canaries in the Coal Mine (Brynjolfsson, Chandar, Chen) finds employment for US workers aged 22-25 in AI-exposed occupations now sits 19% below less-exposed peers, up from 15% a year earlier. The defense is what Naval Ravikant calls specific knowledge: tacit skill built through apprenticeship and real reps, which the data shows AI is not yet replacing.

I read the revised Stanford paper twice this week, because the headline number moved in the wrong direction. Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen released the August 2026 update of "Canaries in the Coal Mine" through the Stanford Digital Economy Lab, built on ADP payroll records. Employment for workers aged 22 to 25 in the most AI-exposed occupations now sits about 19% below where it would be had it kept pace with their less-exposed peers. When the team first documented this gap in August 2025, it stood at 15%. With data now running through June 2026, it is 19% and still widening.

Three details matter more than the headline. The damage runs through reduced hiring, not firings — companies are not cutting young analysts, they are quietly not hiring the next class. Experienced workers show no comparable gap; in occupations where AI complements people rather than automates their tasks, employment is flat or rising. And the revision adds a mechanism I have not seen named this cleanly in labor data before: the split between codified and tacit knowledge. Roles built on codified knowledge — the kind you can absorb from textbooks and documented procedures — are shrinking for the young. Roles built on tacit knowledge, acquired through practice, mentorship, and repeated contact with messy reality, are holding or growing.

That split is not new to me as a coach. Naval Ravikant has been drawing it for years in The Almanack of Naval Ravikant: specific knowledge is "knowledge you can't be trained for," learned through apprenticeships rather than schools, and it feels like play to you while looking like work to everyone else. In 2020 that read as career philosophy. In August 2026 it reads as a forecast that payroll data just confirmed. Large language models are extraordinary at reproducing knowledge already encoded in text. They remain mediocre at the judgment you build by watching a negotiation collapse or shipping a product to real customers. The market has noticed.

Honesty requires the caveats, and the authors supply them. These are descriptive patterns, not causal estimates: the gap shrinks when education is controlled for, and some divergence predates generative AI. The wider productivity picture is murky too. An Atlanta Fed survey cited in an August 2026 TechXplore analysis found roughly 90% of executives believe AI has not yet boosted their company's productivity, and the same piece reports new research showing AI-justified layoffs destroy the employee sentiment that AI-driven gains depend on. Meanwhile Stripe's economics team points the other way, citing task-level studies of 12-14% productivity gains and US labor productivity growing near 2.5% over the past year. Both can be true at once: AI is eating the codified bottom rungs of career ladders while delivering unevenly everywhere else.

So what do you do with this if you are 24 and ambitious — or a founder deciding whether to hire someone who is? In coaching sessions I now run a four-step exercise I call the Tacit Audit. Step one: list the parts of your role a competent stranger could perform from your written notes alone, and assume a model can eventually do those too. Step two: mark where in your week you touch un-codified reality — live clients, broken systems, negotiations, production incidents. Step three: deliberately trade codified output for tacit exposure, volunteering for the sales call or the deployment nobody wants, because that experience never made it into any training corpus. Step four: put your name on outcomes — Naval's accountability principle — so the tacit knowledge compounds to you rather than to a process document.

For founders the implication inverts. If you stop hiring juniors because a model writes better first drafts, you are strip-mining your own future supply of the experienced, tacit-knowledge people the data says survive. The honest opportunity is uncomfortable to say out loud: capable 23-year-olds are underpriced right now, and the ones who will choose tacit exposure over a polished job description are precisely the ones worth taking. Apprenticeship is becoming the scarce asset on both sides of the table.

Sources: Stanford Digital Economy Lab — Canaries in the Coal Mine, August 2026 revision, TechXplore — Layoffs tied to AI hurt worker productivity (August 2026), Stripe Economics — AI and productivity.