AI/ ai · multiagent · research · protocol-design

Langshaw Proposes a Cleaner Protocol for Coordinating AI Agents

A new declarative language gives multiagent AI systems explicit rules about authority and conflict, rather than locking down every step of execution.

Researchers have published a formal protocol language for multiagent systems that tries to solve a coordination problem most agent frameworks currently paper over.

The paper introduces Langshaw, a declarative language built on two concepts. The first is "sayso" - a construct that defines which participant has priority for setting any given attribute in a shared interaction. The second is a pair of conflict constructs, "nono" and "nogo," that explicitly represent when actions clash. The authors also provide formal procedures for verifying whether a protocol is safe and live, and a method for compiling it down to a message-passing protocol that can run asynchronously.

Most multiagent coordination today is either hard-wired or improvised. Langshaw's split between who has authority and what actions conflict gives protocol designers a way to express intent without prescribing a single execution path. As agentic AI pipelines grow more complex - stacking tools, models, and automated decision-makers on top of each other - the absence of a shared formal language for expressing who is in charge of what is a real and underappreciated gap.

This is a research preprint; the distance from a formal semantics paper to tooling that actually ships in production has historically been wide.

TR

The Revision

Written by an AI system from the public sources credited above. How we write →