Respan

Respan initially targeted AI startups building agent-based applications, assuming they would urgently need observability and evaluation tools as their systems scaled into production. The company positioned itself as essential infrastructure for teams struggling with opaque agent behavior and unpredictable failures. However, the available information doesn't provide detailed specifics about whether Respan discovered a different primary audience than anticipated or how their early outreach efforts performed. What we know is that Respan gained traction with over 100 AI startups, suggesting their core assumption—that AI teams need unified control planes for agent monitoring—held sufficient validity to gain early adoption. The fact that they evolved from Keywords AI into a more focused observability platform indicates they refined their positioning based on market feedback. The combination of production observability, automated evaluations, and an adaptive gateway suggests they validated that teams wanted integrated solutions rather than point tools, though the specific customer discovery process and whether initial targeting assumptions required major pivots remains undocumented.

Respan launched their MVP as a lightweight tracing layer that captured agent execution logs without requiring invasive code changes. They shipped within weeks by deliberately excluding the evaluation and gateway components, focusing solely on making observability accessible to teams running early-stage AI agents. This constraint forced them to nail the core problem: visibility into what agents actually do in production. Early validation came quickly—teams immediately started integrating the tracer because debugging agent behavior was painful and manual. The stripped-down approach meant faster onboarding and clearer product-market fit signals. However, this narrowness also limited their TAM initially; customers couldn't fully automate issue detection or remediation without the missing pieces. Once they validated that teams would pay for observability alone, Respan rapidly layered in evaluations and the adaptive gateway. This phased execution prevented them from over-engineering features nobody needed while maintaining momentum with real users who desperately needed *something* to understand their agents' behavior.