The Attribution Wall Is Cracking — But Our Forecast Isn't Resolved Yet

Let's be direct about the resolution criteria first, because the editorial review flags on our last draft caught a real problem. Our forecast resolves on a 'major layoff wave explicitly attributed to AI automation' — and we mean publicly attributed, in official communications, not in survey responses to researchers. The CEOWORLD data is powerful, but it comes from employer self-reporting to a survey instrument, not from SEC filings, press releases, earnings call language, or formal workforce reduction notices. That distinction is not a technicality. It is the entire variable our thesis is built around: that companies are performing AI attribution privately while avoiding it publicly, and that the attribution wall will eventually crack under economic and investor pressure.

So what does the CEOWORLD data actually prove? It's the strongest proximate signal we've seen that corporate willingness to attribute cuts to AI is shifting internally. Forty percent of roughly 80,000 tech layoffs in 2026 labeled AI-caused — that is not a rounding error. Fifteen thousand cuts tied to AI in March alone suggests this isn't a handful of firms experimenting with the framing; it's becoming a normalized internal rationale. This moves our estimate of employer attribution willingness significantly. It does not yet move our estimate of official public attribution, because survey self-reporting and press release language are governed by different incentives. Internally, AI framing is clean — it signals modernization to employees and management alike. Publicly, in a filing or announcement, it triggers scrutiny: labor relations risk, potential discrimination claims, investor questions about whether the cuts were really necessary or just efficiency theater. That gap between internal and public attribution is exactly why our forecast hasn't resolved, and it's why we moved 3 points rather than 10.

We also want to name something our previous draft missed: the attribution bias could cut both ways. We've focused on the undercounting risk — companies avoiding AI attribution publicly to limit legal exposure. But there's a legitimate overcounting risk too: companies may over-attribute to AI in surveys because the 'AI efficiency' framing is more palatable to investors and boards than 'we overhired in 2021 and are correcting.' If survey respondents are strategically reaching for the AI explanation when the underlying driver is balance sheet repair, the 40% figure overstates actual AI causation. We don't know which bias dominates, and intellectual honesty requires us to say so.

The Yale Insights data on junior developer employment is real and concerning, but we're being more careful about it than our last draft was. Employment among developers aged 22–25 is down 20% from its late-2022 peak, and software development postings are down 53% from that same point. Our previous framing asserted this was 'hard to explain as cyclical' — but 2022 was the peak of one of the most anomalous hiring booms in tech history, followed by a well-documented correction in which Meta, Amazon, Google, and Microsoft collectively cut over 200,000 jobs in 2022–2023 alone. A 20% employment drop from that peak is at least partially cyclical normalization from an aberrant high. The structural AI displacement argument is still plausible — and the 53% posting collapse is steeper than we'd expect from pure cyclical correction — but we're not treating it as confirmed structural displacement without more granular data showing AI-substitutable roles declined faster than non-substitutable ones. What would move us above 80%: a major employer's public earnings call or press release using AI displacement as the explicit justification for a reduction in force affecting 1,000+ workers. What drops us below 55%: three consecutive quarters of junior tech employment stabilization, suggesting the 2022–2023 overhang, not AI substitution, drove the decline.