Engineering Founders

The episode contains genuine nuggets - particularly the nuanced argument about betting on model directionality (not just 'models will improve' but specifically 'models will become more agentic'), the closed-loop vs open-loop distinction for coding vs production debugging, and the token-cost consolidation problem for production agents. However, a significant portion is occupied by standard startup platitudes around co-founder trust, hiring, customer focus, and YC endorsements that add little for experienced operators.

“it's less that betting that the model will get better. It's an obvious bet bet. But in what way does the model get better? It's a little bit more nuanced”

“Coding, it's a closed loop thing. You can write, test it can validate it. It's all, it's a closed loop. You can boot up the machine on the production side it's very open ended”

The insight that experienced engineering managers are better at orchestrating AI agents because they already think in terms of delegation and work decomposition is a genuinely fresh and underexplored angle. However, the rest of the episode recycles well-worn startup wisdom - trust your co-founder, focus on customers after pivots, land-and-expand in enterprise - without meaningful reframing or counterintuitive takes.

“people that are able to really break down tasks in a cohesive way will get the maximum mileage out of agen”

“if you've managed teams before, even interns before, like just have thought about delegating work. That helps a lot with managing agents”

The guest is a genuine multi-cycle practitioner: early Facebook engineer who built the first React app and Facebook Business Manager from scratch, founder of an esports startup acquired by Niantic, led Pokemon Go platform infrastructure, and now building an AI-native B2B product. His co-founder's 9-year Databricks tenure including leading the AI agents framework adds further depth. Not a thought-leader - actual operators who have done this at scale.

“One of the first engineers on the Ads team. The claim to fame. We built the very first React app”

“He spent nine years after graduating college at databricks and he managed the infrastructure and enterprise teams. Uh, and the last two years he was there, he became the tech lead for their AI agents framework”

The episode delivers meaningful concrete detail: eval improvement from 3:1 to 30:1 thumbs-up ratio nearly overnight with a model release, ~$1M precede from Afore Capital, Series A from Excel before YC demo day, a Blizzard cease-and-desist that killed the product, 8 staff-to-principal engineers with zero management overhead, and 40 companies approached in one month. Revenue figures, customer names, and ARR are entirely absent, which caps the score.

“we saw our evals go from 3 thumbs up to 1 thumbs down to 30 thumbs up to 1 thumbs down almost overnight with the model shipping”

“we raised a little, I think about a million dollars at the time...we applied to YC and got in...during YC we uh, actually raised our Series A from Excel, um, before demo day”

The host largely plays narrator, stringing together sequential 'and what happened next?' prompts and frequently echoing or summarising the guest's own words back to them rather than probing. There are occasional moments of useful clarification and one genuine follow-up ('But what insight led you to bet on that?'), but no pushback, no challenged claims, and no productive disagreement throughout the 63 minutes.

“And what happened next?”

“And what happened after that.”