Kent talks with Dillon Mulroy, Principal Engineer at Cloudflare, about Agent Experience and dogfooding AI platform work: how Cloudflare closes the loop between builders and customers, why observability and support are product superpowers, and how to stay disciplined when agents tempt you to ship huge diffs overnight.
They go deep on watching users work, firehose social feedback, partnering with support, and why "make pain painful" aligns incentives for better software.
Dillon's path runs from internal insurance tools to Vercel Domains to Cloudflare's agent and dashboard work — always with the same through-line: care about the user, get real feedback, and invest in primitives so delighters don't collapse under bad foundations. This episode covers metrics and paging as a product habit, learning from customer escalations, scoping small when AI speeds up coding, and building cross-functional relationships (support, sales, finance) as part of engineering judgment.
You'll hear practical parallels with episodes on delighters and onboarding tension, plus why reviewing agent-written code still matters for system intuition when things break at 2 a.m.
