Show HN: Product analytics (and evals) for agent sessions on your MCP

Hacker News reports that Armature, a Y Combinator startup, is launching analytics for AI agent sessions. When agents use your tools inside Claude or ChatGPT, you normally can't see what they're actually doing or why they might be failing. Armature reconstructs each session: what the user asked, what the agent thought, and which use cases work best. Integrate in three lines of code. The team is also building an automated feedback loop that identifies common issues, generates fixes, tests them across all models, and opens pull requests to ship improvements. One example from their launch: a marketing platform caught a model hallucination that could have sent campaigns to all their contacts. As more products integrate AI agents, the need to understand what they're doing—and catch hallucinations before they cause damage—is becoming clear.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton