Lemma raises $2.3M to catch silent agent failures

Lemma raised $2.3 million in pre-seed funding and says it now analyzes more than 1 million agent traces per day for loops, failed tool calls, and misread intent. The funding announcement describes a monitoring loop that proposes fixes, while Lemma’s production-observability product turns recurring failures into issues and online evaluations. Teams moving beyond prompt-level controls should instrument semantic task success—not just crashes—alongside the hard-deny controls for coding agents.