AI startup Lemma has raised $2.3 million in a pre-seed funding round as it works to address a growing problem for companies deploying AI agents in production: failures that happen without triggering traditional error alerts.
Founded by Jerry Zhang and Cole Gawin, Lemma has developed a monitoring and observability platform designed to identify when AI agents appear to be working but fail to complete their intended tasks. The startup was recently named by Forbes as one of the top startups to watch from Y Combinator’s Fall 2025 batch.
READ: AI, robotics, and more: YC W26 Demo Day highlights rising influence of Indian American entrepreneurs (March 31, 2026)
Unlike conventional software failures, AI agent problems can be difficult to detect. An agent may complete a task without crashing or generating an obvious error while still misunderstanding a user’s request, making a failed tool call or getting trapped in a loop. These silent failures can go unnoticed until they lead to customer frustration or lost business.
Lemma analyzes live production traffic to identify these issues and trace them back to their underlying causes. The company said its platform is now monitoring more than 1 million agent traces every day. It also proposes fixes and pushes them back into the codebase, bringing detection and resolution into a single workflow.
“Cole and I started Lemma because we experienced the pain of building AI agents firsthand,” Zhang said. “We kept running into the same problem: agents would appear to work, but the results weren’t reliable enough in production.”
“We wanted to build the tools we wished we had: something that helps teams catch issues earlier, learn from production data, and continuously improve agent performance in the real world,” he added.
The need for such tools is becoming more significant as AI agents move beyond experimental use and begin handling critical work across industries including healthcare, finance and law. A failure that does not trigger an error can still affect a customer’s experience, making it harder for engineering teams to identify what went wrong.
Matrix General Partner Ilya Sukhar said Lemma is addressing an early but important challenge in improving the quality of AI agents.
“We’re in the very early innings of improving agent quality, and I’m excited to see the Lemma founders tackle it with a focus on silent failures and automated resolution,” Sukhar said.
Ashley Smith, founder of Vermilion Cliffs Ventures, said existing monitoring systems often focus on identifying technical breakdowns after they occur, while Lemma is focused on determining whether an AI agent actually completed its intended job.
“Existing tools tell you what broke after the fact. Lemma tells you whether the agent actually did its job, catching silent failures before users churn,” Smith said.
The funding round included Matrix, Y Combinator, Liquid 2 Ventures, Vermilion Cliffs Ventures, Irregular Expressions, Cervin Ventures, Comma Capital, Position Ventures and Eight Capital. Angels and operators from companies including OpenAI, xAI, Meta and DoorDash also participated.
READ: Altman’s $2 million OpenAI ‘tokenmaxxing’ offer to startups raises founder control concerns (May 21, 2026)
Lemma plans to use the new funding to expand its product and develop its AI agent failure detection capabilities. The company is targeting teams from the seed stage through Series B that are already operating AI agents at significant production volumes.
Zhang and Gawin first met as freshmen in USC’s startup incubator before eventually founding Lemma. The company is positioning its platform around the idea that AI agents should not simply be monitored for whether they crash, but for whether they actually deliver the right outcome.
As businesses increasingly rely on AI agents to perform real-world tasks, that distinction could become an important part of keeping automated systems reliable.