The work behind the prompts
Share of classified usage by major occupation group · %
Loading chartSelected groups · different platform populations
AI CODING · STACK OVERFLOW 2025
Using AI is not the same as handing over the code
Survey: 29 May–23 June 2025 · reviewed 2026-10-23ANTHROPIC ECONOMIC INDEX
What people actually do with Claude
Loading chartFrom answers to things people make
Classified output types · share of usage · %
Loading chartClassifier estimates · categories can overlap
DEVELOPER CHOICE · 2025Wanted by many
Wanted by many
Chosen again by whom?
Want to useWant to keep using
| Model family | Want to use % | Want to keep using % |
|---|---|---|
| OpenAI GPT | 51.2 | 61.2 |
| Claude Sonnet | 33.3 | 67.5 |
| OpenAI Reasoning | 25.9 | 63.6 |
| Gemini Flash | 24 | 56.6 |
| Gemini Reasoning | 22.7 | 65.2 |
| OpenAI Image | 18.7 | 59.4 |
| DeepSeek Reasoning | 17.1 | 51.5 |
| Meta Llama | 12.5 | 48.3 |
| DeepSeek General | 11.4 | 51.2 |
| X Grok | 8.9 | 52 |
| Mistral | 8 | 49.6 |
| Perplexity Sonar | 6.3 | 54.7 |
| Alibaba Qwen | 4.3 | 53.8 |
| Microsoft Phi-4 models | 4 | 42.6 |
| Cohere: Command A | 1.1 | 41.7 |
| Reka (Flash 3 or other Reka models) | 0.9 | 61.4 |
Stack Overflow 2025 · ODbL ↗n=16,474
Want to use: all respondents. Keep using: existing users. Self-selected 2025 survey · catalogue-covered families.
AI assistance is not the same as vibe coding
Loading chartDifferent question populations · adoption includes plans
More distrust than trust
Loading chartSelected answers · omitted responses are not zero
Almost right still needs work
Loading chartOverlapping self-reports · not measured debugging time
Expected versus measured task-time change
% change in completion time
Loading chartHistorical tools and a narrow sample; the reported 95% confidence interval is +2% to +39% task time. METR's February 2026 update warns that later experiments suffer from selection and time-measurement problems; this is not an estimate of current tool performance. Follow-up estimates are affected by selection and measurement bias; no reliable current effect size established.
