The research out of the Acemoglu-Restrepo extension is the most honest mapping of AI displacement we have seen, precisely because it refuses to sensationalize. The finding that software engineers rank lower than credit analysts and judges on exposure is counterintuitive, but it holds up under scrutiny. The reason is straightforward: agentic systems now chain tool calls and self-correct, which means they can absorb the end-to-end workflow that a senior engineer coordinates. What they still struggle with is regulatory accountability and exception handling, which is why roles with high human stakes or interpersonal overhead remain comparatively insulated. For your organization, this is not a warning to fire your engineers; it is a signal to audit which workflows involve judgment calls that cannot be automated without a human signature.
The 2-3 year adoption lag between metros is where this gets practical. Seattle in 2027 looks like NYC in 2029, and that is not a trivial detail. It means your hiring plans, training budgets, and even office location strategy should be built on a rolling forecast, not a static assessment. If you are in SF Bay, you are already feeling the moderate-displacement threshold that 93% of information-work occupations will cross by 2030. But note what is missing: no occupation hits the high-risk threshold even by 2030. The authors predict widespread moderate exposure, not catastrophic elimination of any single role. That is a different problem than the headlines suggest, and it demands a different response. You are not bracing for a cliff; you are preparing for a slow, broad erosion of routine tasks within every job.
The validation work here is refreshingly transparent. The correlation with the AIOE index at rho = 0.84 and with GPT exposure at rho = 0.72 confirms the signal is not a calibration artifact. But the null result on the 2023-24 OEWS data, reported openly with a falsifiable prediction for May 2025, is the kind of intellectual honesty that should be the standard for this field. It also tells you what to watch: when the 2025 data drops, check whether the correlation turns negative. If it does, the framework is not just academically sound; it is a practical tool for your own workforce planning. If it does not, you will have wasted nothing, because the COV rubric, the part the authors admit is weakest, is the one thing you should treat as a starting point, not a conclusion.
What this means for you is simple. Stop asking which jobs will disappear and start asking which tasks within your team can be chained end-to-end by an agentic system today. The 17 emerging job categories, like AI Reviewers with no coding requirement, are already hiring at real scale. That is not a prediction; it is a job posting. The authors are upfront that their keyword-based COV scores likely underestimate displacement risk by 15-25% for roles with high interpersonal overhead, so err on the side of redesigning roles that involve heavy exception handling. The future is not a binary of safe and doomed. It is a spectrum of workflow coverage, and the organizations that map their own workflows against this framework now will be the ones that absorb the 2030 threshold without a crisis.