The AI & Work Brief: changing tasks and the jobs picture

This week: tasks, opportunities and evidence

This edition covers November 2–8 through 10:17 Singapore time on November 8. It brings together our reporting on training routes, workplace surveys, scientific AI use, announced cuts and official employment figures. It adds November 4 ADP and November 5 Bank of England context. A career decision calls for two questions. What work would you do, and what evidence describes opportunities to do it?

Training routes: a classification of potential

CBI Economics maps about 96% of green apprenticeship participation to occupations with lower AI substitution potential, versus 73% overall. The comparison covers England’s 16–24-year-olds in 2024/25 and includes continuing learners. Lower potential includes low exposure and high exposure where AI may support workers. The report uses older capability mappings, splits multi-occupation standards equally and excludes robotics and autonomous general AI. An annex table note says starts, while the formal methods specify participation. Investigating placements and job security requires additional evidence.

Sources: CBI Economics — Green Advantage; CBI Economics — Technical annex

Workplace surveys: speed and enjoyment

In Miro’s survey, 73% of 9,502 respondents reported better output speed, 56% better quality and 40% better collaboration in the same question. The online survey ran in September–October across 11 geographic samples. Participants were offered an incentive, and the sample split roughly evenly overall between individual contributors and director-level or higher staff. Sampling-frame and weighting details were absent. Respondents supplied their assessments of change.

Atlassian surveyed 1,013 full-time US knowledge workers on September 28–October 1. AI-role questions went to 949 AI users. Separately, 82% said AI took on at least some enjoyable work and 82% reported this for disliked work. The overlap between those groups is unresolved. The article supplies no full questionnaire, sampling frame or weights. Employment effects require separate evidence.

Sources: Miro — The State of Intelligent Collaboration; Miro — Research launch; Atlassian

Scientific work: usage has its own denominator

Anthropic classified 207,049 consumer-plan sessions as scientific research from August 2–September 30. After sampling weights, data analysis represented 33.7% of Chat and Cowork research sessions and 48.1% of Claude Code research sessions. The sample covers Free, Pro and Max, with interactive Claude Code only. Models classified fields and tasks; the main analysis required both. Session shares describe use of Claude, without measuring scientific output, worker productivity or employment.

Sources: Anthropic — Research-session study; Anthropic — Methods appendix

Announced cuts: stated reasons and positions

Challenger recorded 43,994 announced US-based employer job cuts in October. Employers cited AI for 1,265 positions, about 2.9% and the firm’s lowest monthly AI count this year. January–October AI-cited cuts totaled 121,401 of 617,189. The unit is positions named in announcements. Completion and causation require separate evidence. Automation and technological updates possibly involving AI are separate categories.

Sources: Challenger, Gray & Christmas

Employment: two different national measures

US nonfarm payrolls increased by 165,000 in October. August–September revisions added 84,000. Payrolls count jobs in an employer survey, while unemployment, little changed at 4.1%, comes from households. Both reference periods include October 12. These figures are seasonally adjusted, and October payrolls are preliminary.

Canadian employment changed little, down 4,700, while the unemployment rate rose 0.2 percentage points to 6.7%. These seasonally adjusted household estimates count people, cover October 11–17 and exclude the territories from national figures. Monthly sampling uncertainty applies.

The week also brought earlier-period turnover data. BLS reported 5.4 million September hires, little changed, and 3.4 million quits, an increase. ADP’s October private-payroll estimate rose 86,000. Its data and methods differ from BLS. These series describe net employment or worker movements. Identifying AI’s contribution requires evidence on adoption and decisions.

Sources: US Bureau of Labor Statistics — Employment Situation, October 2026; Statistics Canada; US Bureau of Labor Statistics — September 2026 labor turnover, November 3 release; ADP

UK context from the Bank of England

The Bank of England’s November 5 report cites firms’ estimates that AI raised productivity by around 0.5% over the past three years. Firms expected AI to add roughly 0.9 percentage points to annual productivity growth over the next three years. Those are respondents’ assessments and expectations. Box D leaves the exact fieldwork and respondent count unstated. The Bank’s business contacts reported less inclination to hire or replace workers, with little effect on firing so far. This is newly discovered context from a report published during the week.

Sources: Bank of England

Our interpretation: make a small evidence sheet

Choose one real opening and one assignment it describes. Write down what you would produce, what you would need to check and which judgment you would explain to a colleague. List the evidence you still need, such as a course syllabus, a conversation about the role or an example of the employer’s work. Keep the unanswered questions visible. We propose this exercise to make a career investigation easier to discuss.

Use the week’s sources to sharpen those questions. An occupational classification can guide an investigation of training. A workplace survey can suggest questions about daily tasks. Employment data can supply context for the location and sector. The next step is to connect that context to the responsibilities of the opening you are considering.