Chris Tate of Vercel has introduced Zero, a new systems programming language created with AI agents as first-class users of the entire toolchain. The project aims to make compiler output and development tooling natively machine-readable so agents can act on diagnostics programmatically.
Key features
-
Structured diagnostics: every compiler error is returned as JSON with stable error codes, line and column locations, and repair metadata. This removes the need for agents to scrape human-oriented error messages with regular expressions.
-
Tiny output footprint: Zero compiles to very small native binaries without a garbage collector, event loop, or hidden runtime overhead — intended for CLI tools and serverless functions.
-
Full CLI toolchain: a single command covers check, build, run, test, format, inspect, dependency graphs and documentation. All outputs are machine-readable, providing agents with everything they need in one place.
-
Human-readable syntax: despite the agent-first approach, the language preserves a clean, readable syntax for human developers. Files use the .0 extension.
Why this matters
Traditional programming languages were designed for human authors and later adapted for automated tooling. Zero flips that model: the compiler and tooling are designed to communicate directly with AI agents. That enables automated agents to interpret errors, apply fixes, run tests and deploy code more reliably and without brittle text parsing.
Status and availability
Zero is still experimental, but its origin at Vercel — rather than a weekend side project — suggests significant backing for the idea. The language is open-source under the Apache 2.0 license and available on GitHub and zerolang.ai.
Short takeaway
Zero proposes a novel, agent-centered approach to language and toolchain design: structured diagnostics, minimal runtime overhead, and a unified, machine-oriented CLI, while keeping the language readable for humans.
