Michigan professor creates Python-like Jac language for AI agents

University of Michigan professor Jason Mars unveiled Jac, a programming language, and Jaseci, a runtime, that treat AI agents as first-class entities. Jac resembles Python, compiles to Python bytecode, JavaScript and native code, uses PyPI and npm libraries, and adds Rust-style borrow checking via LLVM.

• Jac integrates Python and TypeScript runtimes and compiles to native code
• LLVM integration adds borrow checking and C/C++ library imports
• Data in Jac is modeled as graphs traversed by agents called walkers
• In a test, an agent built a parallel app in Jac but a linear one in LangGraph