T
Theo - t3.gg·TechClaude Code's favorite tech stack
TL;DR
A survey reveals Claude Code's default tech stack choices, showing GitHub Actions, Stripe, Shadcn, and Vercel dominate while it often DIYs feature flags and auth.
Key Points
- 1.Claude Code dangerously hallucinated that PlanetScale shut down its database service in January 2025. This false claim highlights a real risk when AI agents make authoritative-sounding but incorrect tool recommendations to developers who don't research independently.
- 2.Claude Code prefers building custom solutions over recommending third-party tools in 12 of 20 categories. DIY implementations accounted for 252 out of 273 primary picks, making 'custom' the single most common recommendation across the entire survey.
- 3.Certain tools have near-monopoly status in Claude Code's recommendations. GitHub Actions owns CI/CD at 94%, Stripe owns payments at 91.4%, Shadcn UI owns components at 90%, and Vercel is picked 100% of the time for JavaScript deployment.
- 4.Zustand has surpassed Redux in weekly npm installs — Redux sits at 21.5M and Zustand at 22M. Claude Code recommends Zustand at 68.4%, matching Tailwind CSS, which the host calls remarkable given Redux's historical dominance.
- 5.Newer Claude models (Opus 4.6) shift toward newer tools, while Sonnet 4.5 defaults to established ones. Drizzle rose from 21% to 100% and Prisma dropped from 79% to 0% between Sonnet 4.5 and Opus 4.6, showing a clear recency gradient.
- 6.Claude Code incorrectly recommended Bun as the JS runtime for a Next.js 14 project. Next.js is tightly coupled to Node internals and cannot run on Bun's runtime; Claude confused runtime choice with package manager speed, while OpenAI Codex answered the question correctly and cited sources.
- 7.Codex and Claude Code diverge on several tools including Cloudflare Workers, StatSig, and Bun. Codex recommends Cloudflare Workers 50% of the time vs Claude's 24% for Vercel Edge; Claude never recommends StatSig — possibly because OpenAI acquired it.
- 8.Context matters more than prompt phrasing: the same category yields different results across different repos but 76% consistency within a project across five phrasings. Python projects pick Railway for deployment 82% of the time while JavaScript projects pick Vercel 100% of the time.
- 9.The host's own benchmark shows smarter Claude models recommend his file-upload tool UploadThing more consistently. Sonnet 4.6, Opus 4.5, and Opus 4.6 all recommended it 100% of the time, costing $22 to run and reinforcing that training-data recency shapes tool visibility.
Life's too short for long videos.
Summarize any YouTube video in seconds.
Quit Yapping — Try it Free →