LangWatch automates testing and repairs for AI agents
LangWatch has launched Langy, an automated AI engineer that turns production failures into tests and proposed code changes for agent systems.
AI-agent teams often know that a system failed in production without having a repeatable way to reproduce the failure and prevent it from returning.
What happened
LangWatch launched Langy, an automated AI engineer inside its testing and observability platform.
Langy analyses production traces, converts product goals into simulated test scenarios, identifies recurring failures and can propose code changes through pull requests.
Human developers remain responsible for reviewing and approving those changes.
The product connects three activities that are often managed separately: observing agent behaviour, building evaluations and changing the underlying system.
Why it matters
Traditional software tests use predictable inputs and expected outputs. AI agents operate through open-ended conversations, external tools and changing contexts, so failures are harder to reproduce.
An automated system that turns real production problems into persistent evaluations could help teams improve reliability faster.
The risk is recursive complexity. An AI tool proposing fixes to another AI system can introduce new errors or optimise for narrow tests while weakening broader performance.
The bigger picture
A new developer-tooling category is forming around the operational lifecycle of agents.
Companies need traces, simulations, evaluations, permission controls and deployment workflows—not only model APIs.
LangWatch is trying to become the loop that connects those components. Commercial adoption has not yet been disclosed, so the key question is whether teams trust automated diagnosis and code suggestions enough to make them part of production engineering.