What AI builders build with.
Which models, frameworks and tools do UK developers build AI projects with?
33% of UK developers building AI name a specific model provider in their projects, against 39% in China (n=9,542).
Sample: 9,542 developers · source: public GitHub profiles · data to 24 September 2026
By model, framework or tool
| Topic | Figure | Sample (developers) | In words |
|---|---|---|---|
| AI agents | 25.1% | 9,542 | AI agents: mentioned in the AI projects of 25.1% of UK developers building AI, against 24.9% in China (n=9,542 UK developers building AI). |
| Anthropic (Claude, Claude Code) | 20.7% | 9,542 | Anthropic (Claude, Claude Code): mentioned in the AI projects of 20.7% of UK developers building AI, against 20.7% in China (n=9,542 UK developers building AI). |
| MCP servers | 18.7% | 9,542 | MCP servers: mentioned in the AI projects of 18.7% of UK developers building AI, against 18% in China (n=9,542 UK developers building AI). |
| AI coding tools | 17.4% | 9,542 | AI coding tools: mentioned in the AI projects of 17.4% of UK developers building AI, against 21.2% in China (n=9,542 UK developers building AI). |
| RAG / vector search | 14.4% | 9,542 | RAG / vector search: mentioned in the AI projects of 14.4% of UK developers building AI, against 10.1% in China (n=9,542 UK developers building AI). |
| OpenAI | 9% | 9,542 | OpenAI: mentioned in the AI projects of 9% of UK developers building AI, against 13.1% in China (n=9,542 UK developers building AI). |
| Hugging Face / transformers | 5.3% | 9,542 | Hugging Face / transformers: mentioned in the AI projects of 5.3% of UK developers building AI, against 4.5% in China (n=9,542 UK developers building AI). |
| LangChain / LangGraph | 4.7% | 9,542 | LangChain / LangGraph: mentioned in the AI projects of 4.7% of UK developers building AI, against 4.3% in China (n=9,542 UK developers building AI). |
| Google Gemini | 4.4% | 9,542 | Google Gemini: mentioned in the AI projects of 4.4% of UK developers building AI, against 5.3% in China (n=9,542 UK developers building AI). |
| Local models (Ollama, llama.cpp) | 3.8% | 9,542 | Local models (Ollama, llama.cpp): mentioned in the AI projects of 3.8% of UK developers building AI, against 2.3% in China (n=9,542 UK developers building AI). |
| Meta Llama | 1.8% | 9,542 | Meta Llama: mentioned in the AI projects of 1.8% of UK developers building AI, against 1.2% in China (n=9,542 UK developers building AI). |
| Chinese models (DeepSeek, Qwen …) | 1.6% | 9,542 | Chinese models (DeepSeek, Qwen …): mentioned in the AI projects of 1.6% of UK developers building AI, against 10.9% in China (n=9,542 UK developers building AI). |
| Mistral | 0.4% | 9,542 | Mistral: mentioned in the AI projects of 0.4% of UK developers building AI, against 0.1% in China (n=9,542 UK developers building AI). |
How this is measured
Base: developers with an AI project (see Developers building AI). Each topic is the % of them whose AI projects (name, description or topics) mention it (DevInsights::AI_STACK); a developer can count under several. value_json.china_pct is the same share among mainland Chinese AI builders. Overall: % naming a specific model provider (OpenAI, Anthropic, Google Gemini, Meta Llama, Mistral or a Chinese model family). Mentions, not usage: a project named "claude-notes" counts for Anthropic. Measured September 2026, most Anthropic mentions are Claude Code skills, plugins and configuration (tooling for the AI coding assistant) rather than applications built on Anthropic's API; OpenAI's include Codex. Source: public GitHub profiles and their public repositories (the teky search dataset; each finding's date is the date the data runs to). Population: GitHub accounts whose stated location we confirm as the UK or mainland China — the geocoder's country is re-checked against the location text (DevLocation), which drops "Cambridge, MA", "Stanford, CA", "Birmingham, AL" and "Philadelphia, PA"-style mis-codes and bare town names shared with a larger place abroad, and Hong Kong, Macau and Taiwan; organisation accounts and profiles with no GitHub account are excluded. It describes developers who publish on GitHub and state where they are, not every developer: it leans towards open-source contributors, academics and students, and Chinese developers also use Gitee, so China is under-represented. n = developers (GitHub accounts) unless the insight says otherwise.
Cite this
Free to quote with a credit and a link.
ProdReady Recruitment (2026). What AI builders build with. Data to 24 September 2026. https://ProdReadyRecruitment.com/research/dev_ai_stack
Need a custom cut?
Journalists and analysts can ask for a breakdown by sector, city, region or language, or the underlying table.
Ask us