Jonathan joins Alex and David to talk about what actually changes when agents get good enough to do the work. He walks through the 60-service monorepo migration that convinced him the ceiling had moved, why "raising ambition" is now a real coaching conversation with his team, and how Block is using linters and LLM-as-judge to keep GraphQL API design honest.

The conversation goes past the usual velocity story into the harder question: engineering is multiplayer, and most AI tooling is still single-player. Jonathan makes the case that the human touch points — design reviews, pairing, cross-functional feedback — are the part you have to deliberately protect, not the part you optimize away. Plus: adversarial code review with mixed models, why he'll let an agent write his code but not his voice, and the trap of low-hanging fruit that doesn't compound.

https://x.com/JWondrusch

https://x.com/davidnoelromas

https://x.com/voidfiles

Karla joins Alex and David from Sydney, where she's been building security-conscious AI workflows at Lorikeet--a customer service automation company where writing code by hand has officially become "artisanal."

https://www.lorikeetcx.ai/

https://x.com/tetrakazi

Dylan Vassallo is an engineering manager at OpenAI leading the People Innovation Labs team — an eng, product, and design team embedded within the People org, building AI-powered tools for recruiting, HR, internal comms, and the employee experience.

In this episode, Dylan talks about what it actually looks like to be an engineering team inside a frontier AI company, how non-engineers are becoming builders, and what the "graduation" process looks like when a prototype needs to become a production system.

Links & References

https://x.com/dylanvee
OpenAI Codex - https://openai.com/codex
Lattice - https://lattice.com — performance management platform mentioned
Linear - https://linear.app — used for team project tracking
Workday - https://workday.com — referenced as system of record