Jacquard is a next-generation, lightweight programming language designed for a new software development environment where AI models write code and human developers review it. While traditional compilers and build tools identify source code like fingerprint recognition, reacting sensitively to even minor cosmetic changes, Jacquard operates like DNA identification, abstracting and identifying only the core structure and logical intent of the code. Based on a static type checker (Checker) written in OCaml, a continuation-passing style (CPS) interpreter, and a C-emitting native backend, this language provides secure isolation and change detection – features most critical in an agent-centric development paradigm – as fundamental specifications within the language itself. In existing programming language environments, large language models (LLMs) faced limitations where simply renaming variables or making harmless code formatting changes would invalidate existing caches and force a complete rebuild and test. Furthermore, code generated by agents always carried the risk of executing malicious code, such as deleting system files in the user's local environment or unauthorized intrusion into external networks. To address these issues, Jacquard is designed to enforce algebraic effects at the function signature level, and implements explicit capability grants through the --allow option at runtime, effectively blocking unauthorized world effects. Moreover, it revolutionizes build caching efficiency through content-addressed definitions, which structurally hash the underlying logic that remains unchanged even when the code is internally refactored. In practical application scenarios, researchers and developers can safely isolate and validate modification scripts generated by dynamic agents. The built-in Warp testing framework performs simulations of algebraic effects through a discrete probabilistic programming library and derives the optimal solution from multiple bug fix patches based on a virtual test pass probability using Bayesian inference. Users can quantitatively secure safety metrics embedded in the canonical diff of line-by-line changes by combining deterministic testing and caching with error injection techniques, without applying the source code variations proposed by the agent to the target environment.

Jacquard is a next-generation, lightweight programming language designed for a new software development environment where AI models write code and human developers review it. While traditional compilers and build tools identify source code like fingerprint recognition, reacting sensitively to even minor cosmetic changes, Jacquard operates like DNA analysis, abstracting and identifying only the core structure and logical intent of the code. It consists of a static type checker (Checker) written in OCaml, a continuation-passing style (CPS) interpreter, and