Baruch Sadogursky

Tessl built its own software factory. For a company working on how AI builds software, that's a reasonable investment. You're probably not Tessl, though. How much of the machinery around your agents should you be building and maintaining yourself?

In this session, Baruch will argue that platform engineering should give teams a way to assemble factories for their own work, drawing on organizational knowledge they don't have to collect again for every agent. We'll look at how a shared context lake feeds governed workflows: a deployment workflow, for example, can use service ownership to route approval to the responsible team.

But teams will also want agents to write and change those workflows. That puts factory construction itself under the platform's governance, including permission to change the steps where humans intervene. We'll examine where teams need that freedom and which security and permission decisions belong outside the factory they're building.

Skills are more than reusable prompts: they are composable units of behaviour. In this workshop, we’ll take a real end-to-end agent workflow and turn it into a working system built from skills. We’ll show how to identify the right behavioural units, turn them into discrete skills, refine and optimise them using Tessl’s review and eval tools, and bundle them into a reusable plugin. Along the way, we’ll explore why this approach is often more robust than relying on a single giant entry prompt, and how modular skills make agent behavior easier to test, improve, and reuse.