Kent talks with Rhys Sullivan about building Executor and thinking like a product engineer in the AI-agent era: how to design the right primitives, why agent experience is becoming its own product surface, and how to keep quality high when shipping has never been easier.
They cover MCP, code mode, approvals, workspace scoping, docs and APIs as user experience, and why slowing down can still be the right move even when agents make speed feel free.
Rhys has an unusually current perspective on product engineering because he is working right at the edge of the agent tooling shift. The conversation starts with his recent work on Vercel Domains and then moves into Executor, where the challenge is no longer just implementing integrations, but choosing the abstractions that make a system composable, safe, and pleasant to use over time.
What makes the episode strong is how often it comes back to product judgment instead of novelty. Rhys and Kent talk about finding the right primitives, observing how other products solve hard UX problems, resisting the urge to ship every request immediately, and building systems that help agents without letting them become dangerously "helpful."
Homework
- Create a dedicated notes channel or system where you save examples of products doing something well.
- Use those notes as reusable product input: when you need to build a flow later, pull the examples back up instead of starting from scratch.
Resources
Guest: Rhys Sullivan
Host: Kent C. Dodds
Video
