16%
Observed AI-related departures
SThree / YouGov · 2026 · five markets
Respondents who saw colleagues leave after AI substantially changed their work. This is not the share of workers who resigned.
Original source ↗Work & skills · Evidence review
How AI is changing UK STEM jobs
AI is changing technical work. Whether it threatens your role depends on the tasks you do, how your employer adopts it and the skills you can demonstrate.
Sources checked ·
Will AI threaten my job? Start with your tasks.
Some tasks can be automated while the job remains. A role may shift towards review and integration; employers may change junior hiring; workers may leave a changed role; or a company may remove posts. These are different outcomes. The surveys below describe adoption, perceptions and skills needs, not your probability of unemployment.
Look for concrete signals: fewer junior openings, tasks reassigned to senior reviewers, new quality targets and whether training time is protected. Ask your manager how success in your role will be measured as AI tools arrive.
What the surveys actually measure
Read the population before the percentage. These are separate surveys, not a shared time series or a ranking of job risk.
19%
Review costs can outweigh savings
SThree / YouGov · 2026 · UK AI users
UK AI users who abandoned a tool because checking its output took longer than it saved. This is a user subgroup, not all UK workers.
Original source ↗57%
Technical AI skills gaps
DSIT · 2025 survey, published 2026 · UK
Surveyed AI-related organisations reporting technical skills gaps. The overall survey has 119 organisations; it does not represent every UK business.
Original source ↗| Measure | Population & timing | What it can tell you |
|---|---|---|
| 16% · colleagues leaving | 4,126 STEM employees across the UK, US, Germany, Netherlands and Japan; online fieldwork 10 July–1 September 2026; announcement 20 October 2026. | Self-reported observations across five markets. Neither a UK-only result nor a redundancy rate. |
| 19% · tools abandoned | UK AI-user subgroup in the same 2026 survey; UK overall sample 885. The linked summary does not give the AI-user subgroup count. | A signal to measure verification time, not proof that all AI tools reduce productivity. |
| 57% · technical skills gaps | DSIT AI Labour Market Survey 2025; 119 AI-developing or AI-using organisations, with question bases varying; published 28 January 2026. | A self-selected AI-related sample, not a national vacancy or unemployment measure. Fieldwork dates are not specified in the cited executive summary. |
| 58% use AI; 18% regularly | IET UK engineering-employer skills findings, published 6 October 2025. The cited release does not specify sample size or fieldwork dates. | Adoption and regular use differ. These figures describe surveyed engineering employers, not individual workers. |
Rows 1–2: SThree; row 3: DSIT summary and methodology; row 4: IET.
Four changes to watch
The findings suggest questions to ask at work. The implications below are editorial interpretation, not measured predictions.
Junior development needs deliberate practice
SThree reports 32% across its five markets saying seniors now do work previously done by juniors. That describes task allocation, not how many junior jobs disappeared. Ask for supervised debugging, experiment design and review work that builds judgement.
Senior work may shift towards assurance
If drafting becomes quicker, checking assumptions, integration and accountability still take time. Treat this as a workflow question: who signs off the output, and is that responsibility reflected in workload and progression?
Count review time in productivity
The UK tool-abandonment finding is a reason to benchmark a complete task. Compare drafting, checking and rework against a manual baseline; a faster first draft alone is not a productivity result.
Skills gaps create learning priorities
DSIT identifies technical gaps among AI-related organisations, while IET distinguishes adoption from routine use. A practical response is to pair domain knowledge with evaluation, data quality and safe integration, then demonstrate a useful result. Neither survey guarantees hiring or a pay rise.
What this means for your kind of work
Editorial task examples and learning suggestions, not occupation-level risk scores. Actual exposure varies by employer, seniority and responsibilities.
| Work area | Tasks AI may assist | Human responsibilities to strengthen | Next learning focus |
|---|---|---|---|
| Software development | Boilerplate code, test drafts and documentation | Architecture, debugging, security and release decisions | Code review, threat modelling and AI-assisted testing |
| Data analysis | SQL drafts, summaries and chart scaffolds | Metric definitions, causal reasoning and stakeholder decisions | Data validation, experiments and reproducible analysis |
| Engineering | Design variants, report drafts and simulation setup | Physical constraints, safety, site context and sign-off | Simulation validation, systems integration and standards |
| Scientific research | Literature triage, coding assistance and draft protocols | Hypotheses, experimental controls, ethics and reproducibility | Source verification, research software and uncertainty |
| Mathematics & modelling | Symbolic suggestions, model code and scenario drafts | Assumptions, proof, calibration and interpretation | Robustness testing, numerical methods and model validation |
Your next 30 / 60 / 90 days
General career guidance from this site. Use approved tools and non-sensitive examples, and agree any workplace experiment with your team.
- 01
By day 30 · map your tasks
Keep a task log for two working weeks. Note time, repetition, judgement and failure consequences. Pick one low-stakes task and record a manual baseline with a clear quality checklist. Ask which skills your manager expects next.
- 02
By day 60 · run a controlled comparison
Repeat that task with an approved AI tool. Log generation, verification and rework time, plus errors and data-handling constraints. Keep manual practice and ask a colleague to review whether quality is comparable.
- 03
By day 90 · show evidence and agree next steps
Prepare an anonymised case study: task, baseline, outcome and limitations. Agree training time, review ownership and progression goals with your manager. Use the case study in applications where permitted; stop using a tool when the full workflow fails your quality or time criteria.
Questions about AI and STEM careers
Sources, method & limitations
A manually maintained evidence snapshot. Dates refer to publication or fieldwork as labelled; findings describe the surveyed populations. No live labour-market feed or individual prediction is used.
Self-report, selection and question wording limit what can be inferred. The IET release omits sample details; DSIT notes self-selection and varying question bases. No causal UK STEM job-loss rate or cross-survey trend is inferred. Career examples and action steps are this site’s editorial guidance.
SThree / YouGov · 2026 report & methodologySThree · 20 October 2026 announcementDSIT · AI Labour Market Survey 2025 summaryDSIT · full report & sample notes (PDF)IET · engineering skills findings, October 2025
Sources checked ·