Jerry Zhang: Your AI Agents Are Silently Failing. Here's How to Fix That. | Ep 1316

In this episode of The Digital Executive Podcast, host Brian Thomas sits down with Jerry Zhang, co-founder of Lemma, to explore one of the most underappreciated problems in enterprise AI deployment — agents that appear to work but silently degrade the moment they hit real-world production. Jerry opens with the founding story: two college friends who met as freshmen at USC's startup incubator, went off separately to build AI agents at companies on opposite ends of the industry — healthcare and chip design — and came back together after realizing they had been solving the exact same frustrating problem. The pain point that crystallized Lemma came from co-founder Cole's internship experience: spending an entire summer manually throwing prompts into Claude over and over again until an eval set passed, a process so repetitive it begged to be automated. That frustration became the first internal prototype of Lemma, built inside the company where Cole was interning. Jerry then describes the critical lesson from their earliest customers: engineers do not care about finding failures, they care about resolving them. The first version of Lemma surfaced issues well but just gave engineers more homework, which they resented. The breakthrough was closing the loop — not just detecting the root cause but automatically proposing a fix and opening a pull request, so the engineer becomes the reviewer rather than the detective. He closes with Lemma's larger vision: deploy once, learn forever. Today's agents do not just plateau in production — they actively degrade through prompt drift and model drift. Lemma is building the observability and self-identification layer that becomes the foundation for truly self-improving agents, where the system can eventually recognize where it is failing, pull the relevant context, and fix itself with minimal human intervention.

If you liked what you heard today, please leave us a review - Apple or Spotify. 

Learn more about your ad choices. Visit megaphone.fm/adchoices