Claude Code's favorite tech stack
Theo runs experiments on which stack Claude Code is most reliable in, and reports the practical implications: stick to mainstream-shaped TypeScript/Next/Postgres if you want fewer agent failures, because the model has clearly been reinforced on those codebases.
Findings
- TS + Next.js + Drizzle/Prisma + Postgres + shadcn: agent one-shots most CRUD and refactors.
- Go and Python (FastAPI/Django): solid, slightly more iterations needed on refactors.
- Rust: works but agent over-uses unsafe and sometimes fails to reason about lifetimes; expect more handholding.
- Niche stacks (Elixir, Gleam, Zig, Odin): noticeably more failures and hallucinated APIs.
Implication
- Choosing a stack is now also choosing how much agent leverage you get. For greenfield work where speed matters, the popular stack is even more dominant than before because the model multiplies it.