Coval raises $28M Series A for voice AI testing platform

Coval is building an increasingly critical but unglamorous layer of the AI stack: tooling to test and monitor agents that talk to humans. As more companies roll out voice bots for support and sales, the risk surface shifts from model errors in isolation to emergent behavior across millions of conversations. Coval’s pitch—treat voice agents like autonomous vehicles that need simulation, telemetry and structured evaluation—is exactly the kind of operational discipline that has been missing from many GenAI pilots.

This doesn’t move the frontiers of reasoning, but it does make it more feasible to put powerful models in front of real customers without flying blind. In the race to AGI, the constraint is increasingly not “can we get a model to do X in the lab?” but “can we run millions of copies of this thing in production without breaking trust or the P&L?” Companies like Coval supply the guardrails and instrumentation to answer that. As agents become more capable and are delegated more autonomy, these platforms may evolve into de facto runtime governance layers that encode safety policies and compliance requirements.

The strategic takeaway: even if AGI remains centralized in a few labs, a large share of value and risk will live in how enterprises deploy swarms of narrow agents. Those swarms will need robust testing and observability infrastructure.