Exclusive: OpenAI and xAI insiders back Lemma’s $2.3M pre-seed to catch AI agents failing in production

Lemma has pulled in $2.3 million in pre-seed funding to go after a stubborn AI-agent problem: failures that don’t announce themselves. The startup’s pitch is simple and a little unsettling. Agents can pass every check, ship, and then drift into real-world failure with no error message, no crash, and no obvious trail for the team to follow. Customers just stop coming back.

That gap is what co-founders Jerry Zhang and Cole Gawin say they ran into while building agents at separate companies. The two met as freshmen in USC’s startup incubator, then each hit the same wall in production: systems that looked solid on paper but weren’t reliable where it mattered. Zhang put it plainly in a quote from the company: they kept seeing agents appear to work while the results stayed shaky once deployed.

Lemma says its platform watches live agent traffic, looks for issues like infinite loops, failed tool calls, and misunderstandings of intent, then identifies the root cause and suggests code fixes. The company says it reviews more than one million agent traces every day. It’s not just watching the smoke; it’s trying to point at the broken wire.

The round drew angel investors from OpenAI, xAI, Meta, and DoorDash, alongside Matrix, Y Combinator, Liquid 2 Ventures, Vermilion Cliffs Ventures, Irregular Expressions, Cervin Ventures, Comma Capital, Position Ventures, and Eight Capital. Lemma was founded in 2025 and is based in San Francisco. The money is meant to sharpen the product and expand detection tools for seed-to-Series B teams running agents at real scale.

This is a crowded market already. Dash0 raised $110 million from Balderton and became a unicorn in March 2026, FriskAI took in $3.6 million from MaC Venture Capital, and Tsuga raised $35 million in Series A funding. But Lemma is trying to sell something a bit different from the usual dashboard-and-scorecard routine: find the silent failure, then help fix it.