Startup Lemma raises $2.3M to catch silent failures in AI agents

### AI Startup Lemma Raises $2.3 Million to Detect Silent AI Agent Failures

AI startup Lemma has raised $2.3 million in a pre-seed funding round as it develops technology designed to address a growing challenge for companies deploying AI agents in real-world environments. Founded by Jerry Zhang and Cole Gawin, Lemma has built a monitoring and observability platform focused on detecting failures that traditional software monitoring systems may miss. The startup was recently recognized by Forbes as one of the companies to watch from Y Combinator’s Fall 2025 batch. Unlike conventional software applications, AI agents can fail without crashing or generating an obvious technical error. An AI agent may complete a task while misunderstanding a user’s request, making an unsuccessful tool call or repeatedly entering a loop. These so-called silent failures can remain hidden until they result in customer frustration, financial losses or other business problems. Lemma is attempting to address this gap by analyzing live production traffic and identifying whether an AI agent has actually completed its intended task. The company said its platform currently monitors more than 1 million agent traces each day. Beyond identifying problems, Lemma’s system is designed to trace failures back to potential causes and recommend solutions. The company can then push proposed fixes back into the codebase, creating a workflow that combines monitoring, diagnosis and resolution. Co-founder Jerry Zhang said the company was created after he and Gawin experienced the challenges of building AI agents themselves. They repeatedly encountered situations where agents appeared to function correctly but produced unreliable results in production environments. Their experience led them to develop tools aimed at helping engineering teams identify problems earlier and continuously improve agent performance. The need for such technology is increasing as companies move AI agents from experimental projects into critical business operations. AI agents are increasingly being tested or deployed in areas such as healthcare, financial services, legal work and customer support. In these environments, a failure that does not generate a traditional software error can still have serious consequences. Lemma investor Ilya Sukhar, a general partner at Matrix, said the industry is still in the early stages of improving AI agent quality. He highlighted Lemma’s focus on silent failures and automated resolution as an important emerging area. Ashley Smith, founder of Vermilion Cliffs Ventures, similarly argued that conventional monitoring tools often concentrate on identifying technical breakdowns after they happen. Lemma instead focuses on determining whether the AI agent actually achieved the outcome it was supposed to deliver. The latest funding round included Matrix, Y Combinator, Liquid 2 Ventures, Vermilion Cliffs Ventures, Irregular Expressions, Cervin Ventures, Comma Capital, Position Ventures and Eight Capital. Several angel investors and technology operators from companies including OpenAI, xAI, Meta and DoorDash also participated in the round. Lemma plans to use the new capital to expand its platform and improve its AI agent failure-detection capabilities. The startup is targeting companies from the seed stage through Series B that already operate AI agents at significant production volumes. The founders first met as freshmen in the University of Southern California’s startup incubator before eventually launching Lemma together. Their platform is built around a simple distinction: an AI system should not be considered successful merely because it avoids crashing. Instead, companies need to know whether the system actually delivered the intended result. That distinction is becoming increasingly important as businesses entrust AI agents with more complex and consequential tasks. Traditional software monitoring can often identify failures through error messages, system crashes or broken services, but AI agents can produce plausible outputs while still failing their objectives. Lemma is seeking to fill that gap by providing companies with greater visibility into what their AI agents are actually doing. The startup’s approach could become increasingly valuable as organizations deploy larger numbers of autonomous systems across their operations. If AI agents become a standard part of business workflows, detecting and correcting silent failures could be as important as preventing traditional technical breakdowns. Lemma’s latest funding gives the company additional resources to develop that technology and compete in the emerging market for AI agent reliability and observability.