

The Web Wasn't Built for Agents
APIs give agents clear responses, structured errors, and idempotency keys. Websites don’t. A click may or may not register. The page can change mid-task, a session can expire, and many actions can’t simply be undone or retried. Mason Williams, a founding engineer at KERNEL, joins Melanie and Cecil to talk about how to make agents use browsers at scale.
They'll cover keeping sessions and authentication alive through long-running tasks, handling actions that aren’t safe to repeat, managing startup time and latency across a fleet of browsers, and using session replays to understand what went wrong. We'll be diving into where browser infrastructure ends, where the agent framework begins, and which responsibilities sit between the two. Come join us!