First Questions
The questions people ask when they arrive here for the first time, answered plainly from the current data — each number carries its source and its period, and each answer links to the page that owns the detail.
Is AI already taking jobs?
We can't answer that with a yes or a no — and no site can, honestly. Nothing in the data establishes that AI caused any job change: the economy has too many simultaneous drivers for clean attribution, and the site's own impact test, which tried to isolate exactly this, came back inconclusive — the comparison design couldn't be trusted, so the verdict was recorded as "no clear effect yet" rather than dressed up as a finding (the impact tests, verdict per decision record DEC-41, October 2026).
What can be said is what the currently measured economy-wide indicators show, and right now they do not show broad labor-market deterioration: the national stress composite reads 14.8 out of 100 (the site's stress gauge, low tier, August 2026), unemployment is 4.1% (BLS, August 2026), the Indeed job-postings index is 102.0 where 100 is the pre-pandemic baseline (Indeed Hiring Lab, September 2026), and the JOLTS layoffs rate is 1.0% (BLS JOLTS, August 2026). That is a description of today's labor market, not proof of absence — a displacement concentrated in specific occupations would not necessarily move these dials yet, which is why the site watches occupation- and place-level measures separately.
Where to look next: National Pulse (the current gauges) · the impact tests (why the verdict is inconclusive) · Glossary (what "stress" and "exposure" mean).
Are businesses actually using AI?
Yes, measurably — and the measurement is direct. In the Census Bureau's biweekly business survey, 23.8% of U.S. businesses reported using AI in at least one business function in the previous two weeks (Census Bureau BTOS, wave ending September 6, 2026). That series has run for 22 waves since it was re-worded in November 2025, and the share has been rising; a separate worker-side survey (the Stanford Real-time Population Survey) agrees on direction, though the two measure different populations and are never spliced into one series (Census Bureau BTOS and Stanford RPS, November 2025 – September 2026; the wording break and the cross-check are described under how adoption is computed).
Usage at the task level shows the same picture from a different angle. 34.6% of U.S. employment sits in occupations in the top exposure quartile — the quarter of occupations where people actually use AI on the largest share of their work tasks (Anthropic Economic Index, release of June 26, 2026). That is measured use of AI on real work, not a model's guess about which jobs AI could reach.
Where to look next: National Pulse (the adoption series) · how adoption is computed · Data Sources.
Are layoffs being attributed to AI?
Some are — employers name AI in a minority of announced cuts, and the share is tracked monthly. 6.5% of announced layoffs were attributed to AI in the most recent reading (Challenger, Gray & Christmas outplacement firm, August 2026). The site's scenario panel treats a single month above 40% as the signal that AI-attributed cuts have become dominant (the scenario threshold table, calibrated to the 2024–2025 range of 15–25%, documented at /api/methodology). Today's reading is far from that level.
Two honest caveats: a company's stated reason for a layoff is its own framing, not an independent finding, and most layoffs are not announced at all — mass layoffs are, via WARN notices, which the site tracks per metro (WARN filings via Big Local News at Stanford and the New York archive, scraped continuously; see how the local context is computed). None of this attributes any layoff to AI — it tracks how often employers themselves make that claim.
Where to look next: National Pulse (layoff context) · Community (WARN notices by metro).
Is AI replacing or assisting workers?
Both patterns are visible in the usage data, and the balance is measured rather than assumed. In the most recent Anthropic Economic Index release, 47.4% of measured usage was augmentation-style — working with AI: learning, iterating, getting feedback — rather than automation-style, directing AI to do the task outright (Anthropic Economic Index, June 2026 release). The site's scenario panel would call augmentation dominant only if that share passed 60% (the scenario threshold table, Anthropic Economic Index series, June 2026). Today it hasn't: measured use leans slightly toward augmentation, and the exposure methodology computes both separately for every occupation rather than blending them.
One distinction does most of the work here: exposure is not displacement. A highly exposed occupation is one where a lot of AI-assisted work happens — it says nothing by itself about whether employment there will fall. The theoretical indices that estimate how exposed a job could be are shown as separate columns next to the observed-usage score on the occupations ranking, never blended into it, because the two often disagree and the disagreement is informative (how exposure is computed).
Where to look next: Occupations (observed vs. theoretical, side by side) · how exposure is computed.
What about young workers trying to enter the workforce?
This is the area where the watched indicators sit highest, so it gets a straight answer: early-career employment in the most AI-exposed occupations has been declining. Employment growth for ages 22–25 in high-exposure occupations reads −3.6% per year (Stanford Digital Economy Lab / ADP "Canaries", August 2026) — inside the 2023–2025 range of −1% to −4% the signal threshold was calibrated against, and well short of the −6% level that would mark acceleration (the scenario threshold table, per /api/methodology).
The recent-graduate unemployment gap is also elevated: recent graduates ages 22–27 have an unemployment rate 1.64 points above the overall rate (NY Fed College Labor Market data, 2026-Q2) — the highest value in the site's series, which reaches back to 1990 (the site's C4 series, 1990-Q1 through 2026-Q2, served at /api/metrics/C4/history) — though still below the 2.5-point level the panel treats as a signal (the scenario threshold table, calibrated to the historical 0.5–1.5-point range). Both readings are consistent with pressure on entry-level hiring in exposed fields; neither shows the economy crossing its signal thresholds, and neither says AI caused it.
Where to look next: National Pulse (the scenario indicator panel) · the scenario tracker.
What about my job or my city?
The data covers all of it: every scored occupation in the federal task list, and 802 metro and micropolitan areas plus every county (Census Bureau CBSA definitions; the site's geography tiers, described on Data Sources). On Occupations you can find a job's measured AI-exposure score — the share of its task time that people actually use AI for — alongside the theoretical indices. On Community you can find a place's exposure, its current stress (unemployment, postings, WARN layoff notices), and its adaptive capacity (education, broadband, income, industry diversity, from Census ACS five-year data).
Read these as measurements, not verdicts. An occupation's exposure score says how much AI-assisted work happens there, not whether the job disappears; a metro's stress score measures the local economy overall, and can move for reasons with nothing to do with AI — a high-stress metro may simply be in a non-AI downturn. Metros are published with their uncertainty intervals rather than as confident point estimates (the exposure methodology, per decision record DEC-22). And the site publishes no occupation or metro forecasts at all.
Where to look next: Occupations · Community · what we don't compute yet.
Can this site predict what happens next?
No — and that is a deliberate design decision, not a gap. No scenario odds or forecasts are published anywhere on the site. The scenario tracker watches eight published indicators against pre-set thresholds, but it shows the evidence, not a verdict: today exactly one indicator has crossed its line — the job postings index at 103.45 (Indeed Hiring Lab postings index, as of September 18, 2026), favoring a gradual transition — and the other seven have not (the live indicator panel, served at /api/scenario, October 2026). When odds are computed internally at all, they start from an equal prior of one-third each (decision record DEC-01, September 28, 2026) precisely so no prior can manufacture a headline.
Two more reasons to be humble: an indicator can cross its threshold for reasons unrelated to AI — a recession, a trade shock — and the whole methodology is being overhauled (started September 2026) and treated as provisional until it is verified. If calibrated, backtested forecasts are ever published, their accuracy will be graded in the open on the Forecast Track Record page. Until then: measured before modeled, and suppressed rather than speculated.
Where to look next: the scenario tracker · what we don't compute yet · Forecast Track Record.