Guests from
A podcast for engineering leaders shipping agents that do real work — across internal operations and customer-facing workflows alike.
Every team shipping agents hits the same wall: the demo worked. Production didn't. Not because the model was wrong — but because production is a different country entirely.
That gap — between "it worked in the sandbox" and "it works for real customers" — is what this show is about.
You've shipped something. It broke in an interesting way. You learned something the documentation didn't tell you.
The first one that made it past staging — what it took to get there, what you had to compromise, and the moment you knew it was actually working in the wild
"Walk me through the moment you knew the agent was actually doing the job in production."When your agent had to cross system boundaries — auth, context windows, tool call limits — and the seams started showing in ways no benchmark caught
"What broke first when the agent hit real users — and what did you build to fix it?"What building, deploying, and operating agents looks like now versus eighteen months ago — and what your team spends Monday morning doing differently
"What does a Monday morning look like for your team now versus before agents were in production?"Everything you didn't know you'd need to solve — observability, trust, fallback logic, user expectations — that only showed up once real traffic hit
"What's the hardest problem you solved between 'the demo works' and 'this is in production'?"The assumptions you'd throw out, the things you'd build differently, the non-negotiables you now ship with every agent — because experience earned them
"What's the one thing you wish someone had told you before you shipped your first agent to production?"Season 1 brought together the people shaping the infrastructure agents run on — from identity standards and OAuth to agentic UI protocols and enterprise workflows.
This is gold for anyone building agents. The emphasis on trust, identity, and state isolation — so often overlooked until it's too late.
That stat really hits — non-interactive usage already being the norm changes how we think about identity completely.
The shift from user-driven actions to system-led workflows is redefining how we think about scale and security.
