Perhaps not Boring Technology after all

5 days ago

LLMs for programming initially favored widely-used languages like Python and JavaScript, but newer models can effectively work with less common or new tools by consuming documentation and existing code examples.
Coding agents can adapt to private or new libraries not in their training data by learning from existing examples and iterating on their output, challenging the assumption that they enforce 'Choose Boring Technology'.
A study on Claude Code revealed a bias towards build-over-buy decisions and a preferred tech stack including GitHub Actions, Stripe, and shadcn/ui, though human choices can diverge from model preferences.
The adoption of 'Skills' mechanisms by coding agents is growing, with projects like Remotion, Supabase, Vercel, and Prisma releasing official skills to facilitate agent integration.

Hasty Briefsbeta