Microsoft study: 20 jobs least likely to be reshaped by AI — blue‑collar and hands‑on healthcare top the list
A Microsoft Research analysis of roughly 200,000 Bing Copilot conversations mapped user tasks to O*NET activities to produce an “AI applicability score,” showing LLM‑style generative AI fits communication and knowledge work far more than hands‑on trades and in‑person care.
Key takeaways
- Microsoft used real Copilot chats to build an “AI applicability score” measuring task overlap with current LLMs — not a forecast of job losses; see the Microsoft Research blog and the research paper.
- Least exposed jobs are physically intensive or require in‑person care — blue‑collar trades and healthcare support top the low‑applicability groups (coverage: Fortune).
- Most exposed jobs involve reading, writing, summarizing, translation and data work — roles in office, sales, writing, and customer support score highest for AI applicability.
What Microsoft measured and why it matters
Microsoft’s team examined what people actually ask generative AI to do by analyzing ~200,000 anonymized Bing Copilot conversations. Researchers mapped the tasks from those chats to standard occupational activities in the U.S. O*NET database to produce an AI applicability score that quantifies how much current LLM‑style AI can assist with the tasks that make up each job. For full methodology, see the Microsoft Research blog, the research paper, and the arXiv preprint.
Microsoft stresses applicability ≠ displacement: the measure shows where LLMs can help with tasks today, not where jobs will necessarily be eliminated.
Most exposed versus least exposed: the broad pattern
Most exposed: Occupation groups built around information, writing and communication rank highest for AI applicability — computer & mathematical work, office & administrative roles, sales, writers, translators and customer support. These roles often include drafting, summarizing and data tasks people already ask Copilot to perform (see Microsoft research; Fortune).
Least exposed: Jobs requiring on‑site physical work, equipment handling or direct personal care show the lowest AI applicability, pointing to blue‑collar trades and lower‑wage healthcare support roles as comparatively less affected by current LLMs (see Microsoft Research blog; Fortune).
Top 10 jobs Microsoft lists as least likely to be reshaped by current LLMs
Fortune reproduced Microsoft’s ten lowest‑applicability occupations as examples of on‑site, equipment‑focused roles that current generative AI cannot directly perform: dredge operators; bridge and lock tenders; water treatment plant and system operators; foundry mold and coremakers; rail‑track laying and maintenance equipment operators; pile‑driver operators; floor sanders and finishers; orderlies (hospital/healthcare support); motorboat operators; logging equipment operators. (Source: Fortune.)
Extending to about 20 safer roles
Microsoft published occupation‑group patterns rather than a public line‑by‑line top‑20. Combining the explicit bottom ten with other low‑applicability groups identified in the paper yields roughly 20 job areas where generative LLMs have the least direct task overlap, including:
- Hands‑on healthcare and care work: orderlies; home health and personal care aides; nursing assistants (see Microsoft; Fortune).
- Blue‑collar and equipment operation jobs: dredge operators; bridge and lock tenders; water treatment plant operators; foundry mold and coremakers; rail‑track equipment operators; pile‑driver operators; floor sanders/finishers; motorboat operators; logging equipment operators (see Microsoft; Fortune).
- Other low‑applicability clusters: construction trades (carpenters, masons, roofers), electricians, plumbers and pipefitters, HVAC and refrigeration mechanics, automotive and industrial mechanics, heavy truck drivers — noting that other automation forms could still affect some of these jobs (see Microsoft Research blog).
Why these patterns matter for workers and employers
The study links higher AI applicability with tasks that involve reading, writing, summarizing, translating and data work. Jobs requiring steady hands, physical presence, tool use or real‑time situational judgment score lower. Microsoft also observed that occupations typically requiring a bachelor’s degree tend to score higher for AI applicability than those that do not, prompting calls for training, augmentation and policy responses so workers can benefit from AI without being left behind (see Microsoft research; Fortune; arXiv).
Reporting caveats and nuance
- The applicability score reflects current Copilot usage — it does not measure all AI forms or future capabilities (Microsoft Research blog).
- Other automation matters: robotics and computer vision could still change construction, transport and manufacturing outcomes even when LLMs are a poor fit today (Microsoft).
- The “about 20” list here is a synthesis that combines Microsoft’s explicit bottom ten with logical extensions of low‑applicability groups; Microsoft did not publish a single public top‑20 safe list (see Microsoft research; Fortune).
Implications for Redding, California
Redding’s economy — rooted in hands‑on skill, equipment work and in‑person care — maps closely to the study’s low AI applicability clusters. Local policy and business leaders could prioritize the following strategies:
- Jobs to protect and grow: Emphasize blue‑collar trades (carpentry, plumbing, HVAC, electrical, heavy equipment operation, logging) in workforce plans to match near‑term labor demand (see Microsoft; Fortune).
- Healthcare jobs: Invest in pipelines for nursing assistants, orderlies and home‑health aides; demographic trends suggest persistent local demand (see Microsoft).
- Training and apprenticeships: Support trade schools, apprenticeships and employer‑led training to protect workers from displacement in high‑applicability sectors (see Microsoft Research blog; Fortune).
- Business and policy responses: Employers should identify where generative AI can augment office and back‑office tasks; public leaders should plan retraining and placement services to balance productivity gains with worker protection (see Microsoft Research blog).
Links to original research and coverage
- Microsoft Research blog: Applicability vs. job displacement
- Microsoft research paper: Working with AI
- arXiv preprint of the Microsoft study
- Fortune summary of Microsoft’s findings
Conclusion
This reporting preserves Microsoft’s core finding: current generative LLMs show strong applicability to communication and knowledge tasks and limited direct overlap with hands‑on trades and in‑person care. For communities like Redding, that suggests practical policies: strengthen trade training, support care workers, and help employers adopt AI where it truly adds value while protecting the local workforce.
