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

• Lemma has raised $2.3 million in pre-seed funding to seek out and repair silent failures in AI brokers.
• Traders from OpenAI, xAI, Meta, and DoorDash joined the spherical, together with Matrix and Y Combinator.
• Lemma evaluations multiple million agent traces each day to determine the principle causes for failures.

Jerry Zhang and Cole Gawin met as freshmen in USC’s startup incubator earlier than constructing brokers at separate firms. Each confronted the identical challenge: brokers handed all exams however failed in real-world use, with no log knowledge indicating the issue.

In these conditions, there aren’t any crashes or error messages. Prospects simply cease coming again, so growth groups haven’t any option to see what went mistaken. Lemma desires to resolve this drawback with its $2.3 million pre-seed funding. 

The spherical included Matrix, Y Combinator, Liquid 2 Ventures, Vermilion Cliffs Ventures, Irregular Expressions, Cervin Ventures, Comma Capital, Place Ventures, and Eight Capital, together with angel traders from OpenAI, xAI, Meta, and DoorDash. 

“Cole and I began Lemma as a result of we skilled the ache of constructing AI brokers firsthand. We saved working into the identical drawback: brokers would seem to work, however the outcomes weren’t dependable sufficient in manufacturing,” stated Zhang, co-founder of Lemma. 

How Lemma works 

Based in 2025 and based mostly in San Francisco, Lemma displays dwell agent visitors to detect points corresponding to infinite loops, failed instrument calls, and misinterpretations of intent. 

The platform evaluations over a million agent traces every day, identifies root causes, and suggests fixes instantly within the codebase, making detection and backbone quicker. Its launch video exhibits this by recreating the viral restaurant scene from Obsession, highlighting how brokers can appear to be they work till one thing goes mistaken.

The market is crowded and altering rapidly. Dash0 became a unicorn in March 2026 after raising $110 million from Balderton, with shoppers like Zalando and Taco Bell. FriskAI raised $3.6 million in pre-seed funding from MaC Venture Capital to observe agent conduct. Tsuga, a Paris-based firm based by former Datadog engineers, raised $35 million in Series A funding to maintain telemetry inside clients’ cloud environments. 

Most rivals supply dashboards and analysis scores, however Lemma stands out by providing an all-in-one resolution that finds and fixes points, serving to groups that don’t have their very own analysis assets.

Traders and subsequent steps

The funding will assist Lemma develop its product and increase its detection instruments, specializing in seed-to-Collection B groups working brokers at actual scale. The AI agent observability market is worth about $0.9 billion in 2026 and will develop to $14 billion by 2035, at an annual development charge of 35.6%.

“We’re within the very early innings of bettering agent high quality, and I’m excited to see the Lemma founders sort out it with a give attention to silent failures and automatic decision,” stated Ilya Sukhar, basic associate at Matrix.

“Present instruments let you know what broke after the actual fact. Lemma tells you whether or not the agent truly did its job, catching silent failures earlier than customers churn. Jerry and Cole hit the identical wall constructing brokers at two unrelated firms, so that they perceive this drawback from the within. They bring about infectious power and obsessive problem-solving to catching the silent failures that erode belief, they usually’re shifting quicker than anybody I’ve seen,” added Ashley Smith, founding father of Vermilion Cliffs Ventures.

Each incumbent racing to bolt an “AI agent” characteristic onto an present dashboard is assuming clients will hold tolerating failures they’ll’t see. Lemma’s founders are relying on clients reaching their restrict first. Whether or not a two-person staff can capitalise on that earlier than Databricks, Cisco, or Datadog construct the identical factor internally is the open query, and doubtless the one Lemma’s subsequent spherical will reply.