The AI & Work Brief: jobs data, robot costs and career choices

What this edition covers

This brief covers publications released September 28–October 4, 2026, and was published on October 5 as a catch-up edition. Our selection brings together five developments that answer different questions about work. Their publication dates do not make their underlying observations contemporaneous.

Employment: a measured outcome, without an AI explanation

The October 2 BLS release reports September payroll growth of 29,000 and unemployment of 4.2%, both characterized as little changed. These seasonally adjusted estimates measure employment conditions; they do not identify AI's contribution. September payroll data are preliminary.

Sources: U.S. Bureau of Labor Statistics

Robotics: capability and cost are separate questions

Anthropic's September 30 research estimates robot capability for tasks representing 34% of US working time under some conditions, but cost competitiveness for only 0.3%. Task ratings, time estimates, and modeled costs underpin these figures; they are not observed job losses or deployment shares.

Sources: Anthropic

Workforce change: distinguish scenarios from responses

McKinsey's September 29 report models roughly 11 million US workers changing occupations during 2025–2035 under its base assumptions. Automation and wider economic changes drive the scenario; the figure does not count completed transitions.

PwC's September 29 release reports rising AI use alongside reduced access to learning resources compared with its previous survey. These are workers' reported experiences. Its May–June 2026 fieldwork precedes this week's publication; the comparison does not show that AI caused learning access to decline.

Sources: McKinsey Global Institute; PwC

A new spotlight on earlier economic research

BEA's October 2 spotlight revisits an August working paper. It compares AI use reported during April 2025–March 2026 with earlier economic outcomes from 2016–2024. Higher-use state-industry groups had stronger output paths after 2020; employment differences were less precise. This retrospective association cannot establish that measured AI use caused earlier growth. The authors also flag small samples, while an industry-only specification did not recover the same output pattern.

Sources: BEA, Survey of Current Business; U.S. Bureau of Economic Analysis

Our interpretation: build the missing connection

Together, these publications suggest a useful editorial question: what would establish the connection between a tool and a change in someone's working life? A convincing workplace account would describe the task before adoption, the change introduced, and what happened afterwards. It would also examine other explanations, such as a change in demand or a reorganization already underway.

We would look for that connection before using a headline to justify a staffing or retraining decision. Consider a hypothetical team whose new assistant prepares first drafts. A manager might report faster production while employees report more checking. Both could be accurate. Deciding what to do next requires understanding the complete assignment and who performs each part, rather than selecting the more convenient description.

A practical agenda for this week

Pick one decision you actually face: a course, an internal move, or a proposed workflow change. Write down the claim supporting it and the evidence still missing. Ask a relevant colleague or employer for one concrete example. Decide what answer would change your plan.

Keep that investigation small enough to finish. A reviewed work sample or a clarified responsibility can be useful progress. These are our proposed actions, not interventions evaluated by the studies above. Revisit your decision when new evidence arrives, including results that challenge your starting assumption.