AI-assisted coding works with obscure and private libraries when given proper context¶
Insight: Contrary to concerns that LLMs push developers toward popular technologies, current coding agents work effectively with obscure or private libraries when given proper context and examples. Earlier LLM versions showed bias toward widely-trained languages, but agents with good harnesses consult existing examples, understand patterns, then iterate and test their own output to fill gaps.
Detail: Willison observes that Skills mechanisms from Remotion, Supabase, Vercel, and Prisma help agents learn proprietary tools. The key distinction is between LLM recommendations (which default to popular tech) and LLM capabilities (which can handle unfamiliar codebases). When prompted to consult documentation via tools like showboat --help, agents successfully consume docs and work with unfamiliar codebases. This counters the "Choose Boring Technology" concern about AI-era tech choices.
Sources
Related: existing entry "How AI IDEs actually work" in external/claude-code.md — COMPLEMENTS