You will be out of work by 2030

What happened

Researchers built a new measurement for how much AI agents threaten occupations by counting entire job workflows they can automate, not just individual tasks. This expands the estimated displacement risk beyond what older task-level analyses showed — credit analysts, judges, and sustainability specialists now score as moderately threatened by 2030 in major US tech hubs.

93.2% occupations crossing moderate displacement risk threshold by 2030

236 occupations analyzed

0.43-0.47 ATE scores for highest-risk occupations

17 new occupational categories identified as benefiting

6 SOC occupation groups studied

5 US tech regions covered

0.35 ATE threshold defined as moderate risk

assumed Prior automation analysis assumed AI would substitute for individual subtasks, leaving the broader job mostly intact.

found Agentic AI systems can execute entire end-to-end workflows autonomously, which means the displacement risk attaches to the whole job, not just pieces of it.

Why this matters

The shift from task-level to workflow-level automation changes what 'displacement risk' actually means. A task-level analysis might say an AI can handle 40% of a lawyer's work; a workflow analysis asks whether an AI can handle the entire client intake, research, drafting, and filing cycle autonomously — which is a different question about whether the job survives at all. The paper identifies 17 new job categories (AI governance roles, human-AI collaboration specialists) that might emerge, but those are concentrated in the same five tech regions where displacement is highest. The real tension: the jobs being created require skills and proximity to AI companies that displaced workers may not have.

Researchers built a formula to decide which jobs AI will eliminate, and the formula decided judges are near the top of the list. The formula was not cross-examined.

Who wins, who loses

who wins Seventeen new AI occupational categories identified by the paper.

who loses Everyone else.

Jargon decoder

ATE score a number the authors calculated for each occupation estimating how much of the full job an autonomous AI agent could do by 2030

Agentic AI AI systems that can plan, take actions, use tools, and complete a whole job from start to finish without a human directing each step

SOC groups Standard Occupational Classification groups — the US government's standard system for categorizing job types

reinstatement effects new jobs created because of the same technology that displaced other jobs

Acemoglu-Restrepo task exposure framework an existing economic model that measures how much automation risk a job carries by looking at which specific tasks in that job a machine could do

The signal

Why this hasn't landed yet

The finding is framed as a scoring exercise, not a data release, and the score is constructed rather than observed, which gives editors and reporters a legitimate reason to treat it as opinion rather than news. The conclusion, that AI will displace most knowledge workers in tech cities by 2030, is also now a crowded claim, and crowded claims are harder to publish as news even when the underlying methodology is novel.

What happens next

Whether the occupations flagged as highest-risk show measurable employment decline, wage compression, or hiring freezes in the identified regions during 2025–2027, compared to the paper's predictions. Because no one knows. When it comes to AI, <i>all we know is that we don't know nothing (</i>Socrates, Operation Ivy).<br><br>Which city gets a municipal UBI pilot program first? Boston, I'll bet.

The catch

The ATE score is not a regression estimate from observed data. It is a composite calculated algorithmically from O*NET task data using adoption parameters the authors calibrated themselves. No one has validated it against actual displacement outcomes because: those outcomes have not happened yet. The 2030 horizon is four years away and every prior automation wave (ATMs, offshoring, robotic process automation) produced displacement timelines that stretched far longer than models predicted. The paper also covers only five tech-heavy cities and six occupation groups, so the 93.2% figure applies to a pre-selected high-risk slice of the labor market, not the economy.

The longer story

The longer arc

Task-based automation exposure frameworks have been a standard tool in labor economics since at least the Autor, Levy, and Murnane work of the early 2000s, which predicted routine task displacement and was largely correct about manufacturing and clerical work over a fifteen-year horizon. This paper argues that prior frameworks underestimate agentic AI because they measure subtasks rather than full workflows. Whether that extension is methodologically valid is the core question, and prior expansions of automation exposure frameworks have tended to overestimate short-run displacement and underestimate long-run restructuring.

Part of a pattern

This is part of a visible wave of academic attempts to quantify AI's labor market impact before that impact has fully materialized, producing a literature that is simultaneously urgent and methodologically contested. Similar composite scoring approaches have appeared in Goldman Sachs sector analyses and various OECD occupation-risk reports over the past two years. The papers tend to agree on direction and disagree sharply on magnitude and timeline.