AI agents can quickly turn from a promising prototype into a stack of hidden complexity: tool wiring, long-running tasks, sandboxing, approvals, tracing, evals, and the question of how to keep everything inspectable as the agent grows. In this live stream, Kewin Wereszczyński will host Artur Morys-Magiera for a developer-focused look at Eve, Vercel’s open-source framework for building durable backend AI agents. Eve takes a filesystem-first approach, where core agent capabilities live in conventional locations so projects are easier to inspect, extend, and operate.

We’ll start with the problems Eve is designed to address, from reducing boilerplate to making agent infrastructure less fragile. Then we’ll break down its core idea: an agent is just a directory containing instructions, tools, skills, channels, schedules, and subagents. That structure gives teams a clearer way to reason about what an agent can do, how it behaves, and where each part of the system belongs.

The session will move from explanation to implementation. Kewin and Artur will build a small agent from scratch, define its behavior, add a basic tool, and write useful instructions. After that, they’ll run it locally with eve dev, interact with it live, inspect what it does, and adjust the setup in real time.

Join us on June 25 at 5 PM CEST to see how Eve approaches agent development and what a filesystem-first framework can bring to teams experimenting with production-ready AI workflows.