Raindrop

Raindrop builds the first monitoring platform designed specifically for AI agents. The product gives AI engineers visibility into agent behaviour in production: it sends alerts when an agent misbehaves and links directly to the offending events, so teams can dig into the underlying conversations and traces rather than guess at what went wrong. Capabilities include automatic issue detection, semantic search across millions of events, and A/B testing for prompts and models.

The company's founders built the platform from firsthand experience of AI product failures in production, and the stated emphasis is on detecting issues before users do. Despite a small, focused team, the platform is already in use at Replit, Framer, Speak and Clay, and its semantic search operates at a scale of millions of events.

Work at Raindrop sits at the intersection of AI observability, tracing, conversational AI analysis and developer tooling, serving AI/LLM applications and product teams. For engineers drawn to hard production problems in a young and fast-moving part of the stack, the attraction is the chance to work close to the metal of agent reliability alongside a small team that has lived the problem it is solving.