
AI in language development and languages for AI

TypeScript, C# and Turbo Pascal with Anders Hejlsberg
Topic 15 of 21
Published May 13, 2026 · Clip 52:07–1:02:47 (10:40)
Anders describes how the TypeScript team uses AI: code review of PRs, implementing or fixing simple issues, porting a backlog of PRs to the new native compiler, and generating test boilerplate—useful for toil but not yet absolving anyone from understanding. He argues programming languages are key to giving AI determinism: AI is stochastic, but applications need reproducible behavior, so asking AI to write a program that computes the answer converts non-determinism into determinism. The language best suited for AI is the one it has seen most in training; TypeScript's mix of annotation-when-needed and inference keeps token count down and guides AI to better programs. He discusses why locality (no globals, clear imports) matters for AI context windows, and how language services and semantic search via LSP—rather than grep—will increasingly help AI validate generated code, possibly with hot servers providing LSP services to AI agents.