Definition
Plain language
A system that uses formal verification to catch bugs in how multiple AI agents coordinate before they ship.
As stated in the literature
An agent-protocol design framework that pairs LLM-drafted PlusCal specifications with TLC model checking, using counterexample traces to iteratively repair coordination logic.
Why it matters: It brings decades-old formal-methods rigor to agent coordination, catching concurrency bugs that no amount of empirical testing would reliably find.
For example, when a multi-agent workflow has a subtle deadlock, TraceFix can draft a TLA+ spec, find a failing trace, and propose a fix to the protocol.
Heard on the show
“… The full title of the paper is "TraceFix: Repairing Agent Coordination Protocols with TLA+ Counterexamples," and the team is out of Rutgers …”Episode 034 — Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool