Jacquard is a new programming language developed by Friend Machine, designed for a future where most code is generated by machine-learning models and reviewed by humans.

Named after the Jacquard machine, the language analyzes existing programming language structures and syntax to create a new system. Key features include explicit visibility of external effects (like filesystem or network access) in function signatures, requiring runtime permission for such actions. Effect interactions can be recorded and replayed for simulation and analysis. Jacquard also employs content-addressed semantic identity, meaning code renames and formatting changes do not necessitate recompilation or retesting.

Accompanying Jacquard is a testing framework called Warp, which integrates replay, results caching, and handler substitution. The language is installable and usable, with documentation available for agents to digest. The project aims to gather feedback on usability and understandability, particularly in agent systems, to address any pain points for both AI and human users.