What question did this study set out to answer?

The aim is to introduce Aricode as a programming language that enables AI agents to directly compile and execute code.

April 16, 2026Open Access

Aricode: The First Native Compilation Target for AI Agents — Direct x86₆4 Code Generation Without Runtime, Linker, or Standard Library

Key Points

The aim is to introduce Aricode as a programming language that enables AI agents to directly compile and execute code.
Developed a compiler to generate x86_64 machine code directly from the abstract syntax tree (AST).
Created standalone ELF binaries without using LLVM, linker, or standard library.
Conducted benchmarks across 14 challenges with 154 binaries and 12 implementations.
Achieved second place overall in benchmarks, only behind hand-optimized NASM assembly.
Produced binaries 72,000x smaller than those generated by Rust and 12,000x smaller than Go.
Demonstrated the ability to generate, compile, and execute code in under 1 millisecond.

Abstract

Every existing AI coding system generates code in languages designed for humans and depends on heavyweight toolchains to compile and execute it. We present Aricode, the first programming language designed explicitly as a native compilation target for AI agents. The compiler generates x86₆4 machine code directly from the AST — no LLVM, no linker, no standard library — producing standalone ELF binaries (199 bytes for Hello World) that an AI agent can generate, compile, and execute in under 1ms. Benchmarks across 14 challenges (154 binaries, 12 implementations) demonstrate second place overall behind hand-optimized NASM assembly, with binaries 72, 000x smaller than Rust and 12, 000x smaller than Go.

AI에게 질문

Bookmark

View Full Paper

Cite This Study

Edwin F. Veliz Jaramillo (Tue,) studied this question.

synapsesocial.com/papers/69e07dc72f7e8953b7cbec28 https://doi.org/https://doi.org/10.5281/zenodo.19582375

AI에게 질문

Bookmark

View Full Paper