We have all seen the demo where one engineer builds an app in ten minutes with AI. I want to talk about the unglamorous part nobody demos. Getting one person fast is easy. Getting reliable, high-quality output that an entire cross-functional team can trust is a completely different game.

This is a hands-on talk. I will show you the actual guides, sensors, specs, and hooks we run with Kiro at the Bundesliga: what worked, what broke, and what you can copy into your own repo the next morning.

No staged demos. In this session, you will:
- learn how to build a real harness with Kiro: feed-forward guides like steering files, skills and specs that shape an agent before it writes code, and feedback sensors like linter integration, dependency rules, tests, and the occasional AI judge that catch it when it drifts. Plus when to reach for a skill, a steering file, a subagent, or a hook.

- Wire agents into a real codebase: managing context across Jira, Confluence, and your repo, writing skills that hold up over time, governing MCP access, and keeping token cost sane as usage grows.

- See how your own work changes: from writing every line to designing the harness and reviewing agent output, plus an honest take on which engineering skills start to matter more.

So, is an AI coding agent a productivity tool for one engineer, or the operating model for a whole team? Come find out, and decide for yourself.