Your software should build itself

Programming agents have a fundamental problem: they're separate from the code they're building. Per LessWrong, a researcher proposes embedding the agent directly into the programming language itself — able to not just use it, but modify it. Max von Hippel calls this an 'Auto-Syntactic Model,' and argues it solves major frustrations with today's coding AI: current systems struggle to edit code precisely, search large codebases efficiently, and often generate wasteful or redundant code. If the agent lived inside the language's structure, it could use type information to surgically edit components, query the codebase hierarchically, and enforce clean patterns automatically. The thinking remains early-stage and speculative, but von Hippel has sketched a minimal prototype and outlined a vision where software increasingly builds itself.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton