AI Usage Peaks in Tech Tasks, Augments 57% of Work
Claude.ai data from 1M conversations shows AI heaviest in software dev (37%) and writing (10%), augments 57% vs automates 43% of tasks, concentrated in mid-high wage jobs like programmers ($75-100k).
AI Concentrates in Specific High-Value Tasks, Sparing Manual Labor
Analysis of ~1M anonymized Claude.ai conversations reveals AI adoption skews heavily toward "computer and mathematical" tasks (37.2% of queries), like code debugging, software modification, and network troubleshooting—far exceeding their 3.4% U.S. workforce share. Arts, design, media, and entertainment follow at 10.3% (vs 1.4% workforce), driven by writing/editing. Lower shares hit office/admin (7.9% vs 12.2% workforce), education/library (9.3%), life sciences (6.4%), and business/financial (5.9%). Physical roles like farming/fishing/forestry register just 0.1%, matching their tiny workforce presence. Depth varies: 36% of occupations use AI in ≥25% of tasks, but only 4% reach ≥75%, indicating partial integration rather than wholesale replacement.
To apply this: Track your own workflows against O*NET's 20k tasks (e.g., prioritize AI for pattern recognition or iterative writing shared across jobs like design and radiology). Overrepresentation signals early movers—software engineers amplify output 11x their workforce proportion.
Augmentation Outpaces Automation in Real Use
AI enhances humans more than replaces them: 57% augmentation (task iteration 31%, learning 23%, validation 3%) vs 43% automation (directive 28%, feedback loop 15%). Augmentation involves collaboration—brainstorming, skill-building, refining—while automation handles direct execution like document formatting. No occupations show full takeover; AI diffuses across tasks.
Practical takeaway: Design prompts for augmentation to boost productivity without displacement risks. E.g., use Claude for iterative code review (augmentation) over one-shot generation (automation), yielding verifiable gains in complex domains.
Mid-to-High Wage Jobs Lead Adoption Due to AI Fit and Access
AI usage correlates with median wages peaking mid-range ($60k-$100k), e.g., programmers/software devs at 3-6% usage ($75-100k). Low-wage manual roles (shampooers ~$25k, <1%) and elite physical/high-skill jobs (obstetricians ~$200k, low usage) lag, reflecting AI limits on dexterity and barriers like training costs. U.S. median wage: $60k.
Implication for builders: Target mid-wage knowledge work for AI pilots—expect 36% task coverage quickly. Low adoption in extremes predicts slower disruption there, buying time for upskilling.
Clio Methodology Enables Privacy-Preserving Task Mapping
Clio (Anthropic's tool) aggregates conversations into O*NET tasks/occupations without exposing raw data, filtering ~1M Free/Pro chats to work-relevant ones. Matches queries to 20k tasks, groups into 6 categories. Open dataset on Hugging Face validates via Appendix B. Caveats: possible hobby use, post-response edits blurring auto/aug, Claude.ai bias toward coding, no image gen. Future: longitudinal tracking of depth, auto/aug ratios to forecast job evolution.