Zero, a new systems programming language designed for AI agents, has shipped with a toolchain built around machine-readable feedback loops.
The compiler emits JSON diagnostics so agents can parse errors directly instead of scraping text, while explicit capabilities make a program's effects visible and constrainable. Typed safe fixes let an agent apply suggested repairs with confidence that they preserve types and behavior. The bet is that languages optimized for human ergonomics leave agents guessing — Zero treats the agent as the primary author and structures every error, capability, and fix to be acted on programmatically.