Inkdown
Start writing

core

6 files·1 subfolder

Shared Workspace

core
Revision w/ Whiteboard

zero-language-explanation

Shared from "core" on Inkdown

Zero Programming Language - Core Working

What is Zero?

Zero is a new programming language created by Vercel, specifically designed for AI agents. It's a compiled language that produces native binaries (like Rust/C++), not an interpreted language like Python or JavaScript.

Key Characteristics

Core Design Principles:

  • Static dispatch
  • Explicit capabilities
  • No mandatory garbage collection
  • No hidden runtime tax
  • JSON-native for structured diagnostics

Architecture Diagram

Plain text
CN Basics - 1
CN Basics - 2
DNS
Event loop
programming-language-concepts.md
zero-language-explanation.md

Why No LLVM? (The Core Trade-off)

LLVM Advantages:

  • Decades of optimization work
  • Automatic multi-architecture support (x86, ARM, etc.)
  • Battle-tested stability
  • Edge case handling

LLVM Disadvantages:

  • Slow compilation time (contributes to Rust's slow builds)
  • Heavy runtime overhead during compilation

Zero's Approach:

  • Custom C compiler written by hand (likely with LLM assistance)
  • Manual binary generation for each CPU architecture
  • Faster compilation, but:
    • Less stable (currently buggy on Mac M1/ARM64)
    • Manual architecture support
    • More maintenance burden

Borrow Checker (Rust-like)

Zero includes a borrow checker similar to Rust (though less powerful):

Plain text

AI-Focused Features

This is Zero's unique selling point - designed for AI agents:

Traditional Compiler Error:

Plain text

Zero AI-Friendly Error:

JSON

AI can easily parse, understand, and fix this

Current State

Bootstrapping Problem:

Mature Languages (GCC, Rust):

Plain text

Zero (Current):

Plain text

Known Issues:

  • Darwin ARM64 (Mac M1) not working
  • Various bugs due to custom compiler implementation
  • Limited architecture support (manual per-arch compilation)

Summary

Zero's Core Innovation: A systems programming language optimized for AI agents through:

  1. JSON-native toolchain output
  2. Structured error codes for LLM consumption
  3. Type repair metadata
  4. AI-friendly diagnostics

Trade-offs:

  • Faster compilation (no LLVM)
  • AI-optimized output
  • Less stable (custom compiler)
  • Manual architecture support
  • Currently buggy

The language is experimental and in early stages, but represents an interesting approach to making programming languages more accessible to AI agents through structured, machine-readable toolchain output.