FriendMachine has officially dropped Jacquard Lang, an open-source programming language purpose-built for a workflow where machine learning models generate code and humans review it. The announcement represents a deliberate pivot away from fully autonomous AI coding toward what the company calls "human-in-the-loop" software development—a framework that treats AI output as drafts rather than finished products.

Why Another Programming Language?

The market already has no shortage of languages, but Jacquard Lang carves out a specific niche: it was architected from the ground up to make AI-generated code reviewable by humans. The language features a compact surface syntax designed to be readable even when the underlying logic comes from an LLM. This isn't about making developers' lives easier in the traditional sense—it's about creating explicit trust boundaries between what machines propose and what humans approve.

Technical Foundation

Jacquard Lang runs on an OCaml-based type checker that validates code before it ever reaches production. The compiler then emits C, giving developers a familiar, battle-tested output target with decades of tooling behind it. This two-stage approach—OCaml checking followed by C emission—means the language inherits ML-friendly semantics while maintaining compatibility with existing C ecosystems and infrastructure.

Rethinking AI Trust Boundaries

The human-in-the-loop design introduces what FriendMachine describes as a new class of trust boundary in software development. Rather than treating AI code generation as an autonomous process, Jacquard Lang forces explicit review checkpoints where humans must validate machine-generated logic. This philosophical stance puts the language firmly at odds with approaches that push for fully automated pipelines from prompt to deployment.

Open Source Strategy

By releasing Jacquard Lang as open source, FriendMachine is betting that community adoption will drive tooling improvements faster than proprietary development could manage. The company has published the compiler toolchain and checker implementation on GitHub, inviting developers to contribute to a language specifically designed for an AI-augmented future—assuming that future still values human oversight.

Key Takeaways

  • Jacquard Lang targets ML-to-human code review workflows explicitly
  • OCaml-based type checker + C-emitting compiler architecture
  • Compact surface syntax prioritizes human readability over machine convenience
  • Human-in-the-loop design creates explicit trust boundaries in AI-assisted development
  • Open source release aims to accelerate tooling and community adoption

The Bottom Line

This is the kind of pragmatic thinking we need more of in AI tooling. Fully autonomous code generation sounds sexy until you remember that LLMs hallucinate, edge cases bite back, and someone has to be accountable when production goes down. Jacquard Lang makes humans stay in the loop—because sometimes the old ways are still the right ways.