US labour market · correlation, not causation

AI can do the tasks. Replacing the jobs is a different equation.

A US-focused model of AI task-exposure, BLS labour-market signals and the full cost of automation. The pattern so far: in most occupations, once you price in supervision, integration and error risk, AI does not yet beat fully-loaded human labour — which is what the macro data shows too. This instrument lets you find where that line actually sits, and move it.

Test the economics Open the signal board

Careful framing: this visualises exposure, labour-market stress and replacement economics — not causal proof that AI created any unemployment change. Data vintage: US unemployment benchmark 4.3% (BLS, May 2026) · wages BLS OOH May 2024 · occupation unemployment CPS 2025 · exposure scores curated from ILO / O*NET / AIOE lineage.

timeline layer

Official labour signal board

This board focuses on US occupation-level signals: BLS CPS 2025 unemployment where available, AI task-exposure scoring, and live scenario-based replacement economics. The chart changes when you select an occupation or adjust the economics controls.

Labour Signal Board

US · 2021–2026 · source-backed labour signals

High pressure Mixed Lower Right column — P pressure · E exposure · U unemployment

Event markers

LLM / coding systems

occupation layer

Exposure × stress × economics

Each bubble is a US occupation family. Right means higher AI task exposure. Up means higher labour stress signal. Bubble size reflects US labour-force scale; BLS CPS/O*NET/BLS wage notes are shown in the occupation panel where available.

Occupation signal map

click a bubble

—

AI exposure

—

Labour stress

—

Task automatable share

—

Reliability risk

—

Human cost / h—

AI-assisted cost / h—

Pressure—

Select an occupation to inspect replacement economics.

financial layer

AI replacement economics

The core test: AI becomes economically disruptive only when capability, reliability, integration, compute and supervision costs beat fully-loaded labour cost.

Scenario controls

change assumptions

Model class QA burden 35% Retry / failure rate 22% Integration overhead 38% Error-cost sensitivity 38% Tasks per hour 30

Economic threshold

selected occupation

Fully-loaded human labour—

AI workflow cost—

Select an occupation.

Replacement Pressure Index Exposure × Task Share × Economic Edge × Reliability Adjustment

watchlist

AI-exposed occupation watchlist

The table prioritises where to look: exposed occupations where labour stress and replacement economics align. Use it as an analytical watchlist, not a verdict.

Occupation	Exposure	Stress	Human cost	AI cost	Pressure

what the data says

Hype outran the economics.

The replacement story is loud. The labour-market evidence, so far, is quiet — and that gap is the whole point of this instrument.

Across these occupations, high AI task exposure rarely turns into a clean economic case for replacement once you add the parts employers underestimate — supervision, integration, retries and the cost of being wrong. Strip those away and AI does start to beat expensive cognitive labour; it still loses to cheap labour, because the workflow has a cost floor a low wage already undercuts. That is a very different picture from "AI replaces everyone."

It also matches the macro record. Independent analyses find little sign of economy-wide disruption yet, while flagging early, uneven pressure on younger and entry-level workers in the most exposed roles. The signal to watch is not aggregate collapse — it is where exposure, labour stress and replacement economics start to line up.

The Budget Lab at Yale & Brookings (Oct 2025) ↗No discernible economy-wide labour-market disruption in the first ~33 months after ChatGPT; no clear link between AI exposure and unemployment.
Budget Lab: "AI is probably not (yet) the reason" (2026) ↗The labour market has cooled, but the data do not yet show clear AI-specific effects once exposed and unexposed occupations are made comparable.
On what we don't know — incl. Brynjolfsson, Chandar & Chen (2025) ↗The clearest early signal is among young / early-career workers in highly AI-exposed jobs — narrow, not yet economy-wide.

Bottom line: the task capability is real and arriving fast; the economic case for wholesale replacement is not here yet, and where it arrives it will arrive unevenly. Worth watching — not panicking.