The AI & Work Brief: adoption, expectations and evidence about jobs

The week's question: what would change your decision?

This edition covers October 12–18 through 09:02 Singapore time on October 18. Our reading of the week's evidence is that adoption, changing responsibilities and employment outcomes need separate questions. Someone choosing a course, applying for a first role or setting a team target needs more than a headline about how quickly AI is advancing.

Support needs: workers and managers describe different problems

Elon's October 13 release found 52% of workers rated guidance on AI's limits and human judgment extremely or very helpful. The worker subset numbered 775 within 1,158 US adults surveyed August 20–24. SSRS used a weighted probability panel with English online and telephone interviews. The finding measures desired support, not whether training improves performance.

Bluebeam's October 13 release addresses an employer-side obstacle: 34% of 448 respondents at AI-using organizations cited inadequate internal AI skills as a scaling barrier. Hanover's July online survey covered 750 US architecture, engineering and construction technology decision-makers at manager level or above. The subset includes pilots and broader use; it does not represent construction workers generally.

Sources: Elon University; SSRS / Elon University; SSRS; Bluebeam; Bluebeam via GlobeNewswire

Use: task shares and text volumes measure different things

IVADO's October 15 announcement highlights a weighted, nonprobability survey of 1,015 Quebec workers, conducted June 26–July 2. Among generative-AI users, 64% reported using it for less than a quarter of their tasks; the user chart's base is 494. That describes reported task coverage, not time saved or jobs automated.

OpenAI's October 14 release reports that Codex and ChatGPT Work supplied 55% of generated text tokens in its EU Enterprise and Business analysis. The trailing 28-day window ends September 30 and excludes Other. Work's July launch changed the product mix. Sample counts and a continuous account cohort are not established; token volume cannot show completed work or productivity.

Sources: IVADO; CRIMT; OpenAI; OpenAI Economic Research

Jobs: a broad tracker and one company's hiring mix

Yale's October 14 tracker incorporates September Current Population Survey data. Occupational composition, employment and wages show no clear broad AI disruption. Relative unemployment growth among exposed workers is elevated in the third quarter, but volatile and uncorroborated by other measures. The comparison uses potential occupational exposure, not verified adoption; aggregate results can conceal narrower changes.

Anthropic's October 16 internal study reports that juniors' share of technical hires fell from 27% to 9% across its 2025–mid-2026 series, while absolute technical hiring grew. Internal records and 17 interviews inform this company case. A precise junior cutoff is unspecified. The hiring pattern does not establish AI causality or a trajectory for other employers.

Sources: The Budget Lab at Yale — Tracking the Impact of AI on the Labor Market; The Budget Lab at Yale — AI Labor Market Tracker: October update; The Budget Lab at Yale — What We Do and Don't Know About How AI is Affecting the Labor Market; Anthropic

Expectations: disagreement is evidence about beliefs

A separate October 16 Anthropic study finds staff expect AI to handle an average 75% of their current work hours by 2030, versus 28% among employed US respondents. Morning Consult's weighted opt-in survey reached 10,980 adults August 11–23; the comparison uses the employed subset asked about 2030. An unweighted internal poll reached 1,334 staff August 9–12. Staff sliders differ from public ranges converted to midpoints; public not-sure answers were excluded, while staff could skip. These are expectations, not measured automation or a company staffing plan.

Sources: Anthropic — The AI expectations gap; Anthropic — Survey appendix

Our practical agenda: keep a short decision record

For the next career or team discussion, write down the decision first. A training choice needs a specific learning objective. A workload change needs evidence about accepted results and review effort. An entry-level hiring decision needs an account of how newcomers will learn. Each requires information that a general adoption statistic cannot supply.

Then record what would make you reconsider. It might be a failed quality check, a new responsibility or a missing opportunity to practice. This is our editorial proposal, not an intervention tested by these studies. Their findings can help frame those questions, but combining them cannot produce a defensible total of jobs gained or lost.