Zero: A Systems Language Designed for AI Toolchain Integration

The Core Innovation

Zero is a systems programming language that prioritizes machine-readable output across its entire toolchain. While most languages generate human-centric text for errors and warnings, Zero provides a native JSON output format for all compiler diagnostics. This allows AI agents to parse error codes, severity levels, and suggested fixes directly from the toolchain without needing to be trained on the language's specific documentation or syntax.

Language Features and Tooling

Zero incorporates explicit capability-based security and memory management primitives similar to Rust or Zig. Key features include:

• World Capability: Functions performing IO operations must explicitly declare the world capability, allowing the compiler to reject invalid IO calls at compile time based on the target (e.g., WebAssembly).
• Error Handling: The language uses a check keyword to propagate errors from functions marked with raises, functioning similarly to the question mark operator in Rust.
• Memory Management: Developers manage memory through mutable spans for writable views and owned types that trigger a drop function upon scope exit, supplemented by a defer keyword for cleanup logic.
• AI-Driven Repair: The zero fix command provides structured JSON diagnostics that include specific repair instructions and safety levels, enabling LLMs to resolve code errors autonomously by iterating through the toolchain output.

Practical Utility

While the language functions as a capable systems language, its primary value proposition is the reduction of friction for AI agents. In testing, an LLM with no prior knowledge of Zero successfully debugged a broken file by relying solely on the structured JSON output provided by the zero fix and zero check commands. Despite this, the language faces the hurdle of competing with established ecosystems like Rust or Go, which already possess vast training datasets for LLMs, rendering the need for a dedicated language for AI-readability an open question.