Why We Invested in ChatSee: Catching What Agents Get Wrong
Chips and models get the press. The next massive AI opportunity is the governance layer quietly deciding whether enterprise agents actually work. True Partner Puneet Agarwal shares why we invested and what it looks like when there’s a behavioral assurance layer protecting performance.
Chips and models are getting all the press. The next massive opportunity in AI, however, is around governance — as boring as that sounds.
Picture the world we're walking into. Thousands of agents, soon tens of thousands, running across multiple models, touching multiple systems, making non-deterministic decisions on behalf of the enterprise. Even at the optimistic end of reported error rates (2 to 5 percent) the operational and security exposure is enormous. A system can be 99.999% available and still be wrong one time in twenty. Traditional observability tells you that something happened. It does not tell you whether what happened was right.
That gap is where ChatSee lives.
ChatSee tracks all agent behaviors, classifies failures, and resolves issues immediately. At its core, the product is a powerful taxonomy and memory map of every agent action across an enterprise, paired with the remediation each failure needs. Detect. Classify. Remember. Correct. Do it inline, at machine speed, across every agent the enterprise runs, whether built in-house or inherited from a vendor, at run-time.
This is the kind of middleware that will quietly determine whether AI in the enterprise actually works. Not the model or the app. The behavioral assurance layer underneath.
We’ve known Sekhar Sarukkai for years. He is a five-time founder and his last company, Skyhigh Networks, sold to McAfee in 2017 in a deal valued close to a billion dollars. Sekhar and his co-founder, Sanjay Agrawal, reconnected with us at Connected Stack last fall, and we jumped at the chance to work with them.
We couldn’t be more excited to be leading their seed round and to be on this journey with them.