Jac
Jac is a Python-like programming language created by University of Michigan professor Jason Mars for developing AI agents, alongside its Jaseci runtime.1 Its central premise is that agentic software needs tools designed for agents rather than workflows shaped around human developers navigating separate languages, libraries, and runtimes.1
Design and capabilities
Jac aims to unify development across Python, TypeScript, JavaScript, C/C++, and native machine code, and can compile to Python bytecode, JavaScript, and machine code.1 It can also use packages from the PyPI and npm ecosystems.1 The language reportedly includes LLVM-based borrow checking and ownership semantics inspired by Rust, intended to address safety and performance.1
Jac provides an “eject” function that can reverse-compile Jac code into TypeScript, Python, or C, which Mars described as useful for compliance or portability needs.1 Its data model is graph-based, representing data with nodes and edges rather than conventional tables or objects.1 Jac’s “walkers” let agents traverse those graphs and perform work in parallel.1
Use and adoption
In a demonstration discussed by The Register, Mars contrasted a parallel application built with Jac with a simpler linear workflow produced using LangGraph; this was his argument for Jac’s capabilities, not an independent benchmark.1 The article reports that Jac is already used in some smaller AI products, while noting that wider adoption is uncertain.1