Michigan professor develops graph-based language for building AI agents
University of Michigan professor Jason Mars has developed Jac, a Python-like programming language, and Jaseci, its runtime, to make AI agents central to how software is built. He argues that tools designed for human programmers leave agents dealing with unnecessary complexity, making code translation costly and agent-driven work less efficient.
Jac combines Python and TypeScript ecosystems, compiles to Python bytecode, JavaScript and native machine code, and can use libraries from PyPI and npm. LLVM integration adds Rust-style ownership and borrow checking and access to C and C++ libraries; an “eject” feature can translate Jac code into Python, TypeScript or C. The language structures core data as graphs that agents can traverse with parallel “walkers”. In a demonstration, an AI agent built a more sophisticated, multi-stage code-review app with Jac than with LangGraph; Java is not currently supported.
In short
- Jac is designed to make AI agents first-class users of a programming language.
- It combines Python and TypeScript ecosystems and represents core data as graphs.
- A demo agent built a more advanced code-review app with Jac than LangGraph.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
A language designed around AI agents could reduce the complexity they face when building software, making agent-driven development more efficient and capable. Jac’s graph structure, parallel walkers, broad library access and multiple compilation targets offer a considered attempt to give agents useful tools while remaining connected to established programming ecosystems.
The case against
A new language and runtime add another layer for developers and organisations to learn, maintain and integrate, while support for familiar ecosystems does not guarantee broad adoption or reliability in practice. The reported code-review demonstration is a limited comparison, and gaps such as the lack of Java support may constrain usefulness for teams with existing systems.
Read the full article at the source →
Originally published by The Register as “Michigan prof creates Python-like Jac language for AI agents”.