Scrum with AI

Deciding, checking and answering for it when building is cheap

A short book, written in the open, about how to use Scrum to adopt AI with sound judgement: in the team that builds software and in the company that adopts it.

AI makes building cheap. It does not make it cheaper to decide what to build, to check that it works or to answer for it. Scrum exists for the second part.

If you used Scrum to protect the capacity of a scarce development team, with AI you no longer need half of it. If you use it to learn fast without losing control, you need it more than ever.

This book explains how to use Scrum to adopt AI with sound judgement and with real effect: in the team that builds software with agents and in the company that adopts its first use cases. It is short, you can read it in an hour and a half, and I am writing it in the open, chapter by chapter.

Contents

  1. Building is no longer the hard part: What the book argues, who it is for and how to read it in an hour and a half.
  2. Scrum is for what you don't know how to do yet: When Scrum helps and when it gets in the way, in software and beyond, and why AI brings more complex work, not less.
  3. The bottleneck always moves: The history of software development told through its value stream, and the answer to those who say Scrum is unnecessary with AI.
  4. Some things cannot be delegated: What each person on the team decides when AI proposes almost everything, and who answers for what it produces.
  5. The Scrum Master doesn't need to master AI: they need to see what it does to the team: The Scrum Master's job with AI: helping the team become aware of how it masters AI, how it organizes itself around it and what problems it brings.
  6. An agent is not a teammate: it is capacity: What work you give an agent, what you don't, and why the spec is the backlog item of the hybrid team.
  7. Learning when AI writes the first draft: How a team keeps learning when it no longer writes the first draft.
  8. Discover and deliver at the same time: Two tracks in the same team and the same Sprint: discover what is worth building and build what is already validated.
  9. Before you build, check that it is needed: What to validate, with whom and how, now that a prototype costs an afternoon.
  10. Planning when effort no longer rules: How to plan and prioritise when AI makes effort cheap and everything looks like a quick win.
  11. Five events, one question: what do we decide here?: What AI does and what the team decides in each Scrum event and artifact.
  12. "Done" is not "the agent has finished": The quality contract of a team that works with AI, and how it changes when the product itself has AI inside.
  13. One ladder of goals: How to link strategy to the Sprint so that AI does not produce a lot of what does not matter.
  14. Look at the whole stream, not your piece: Where AI speeds things up and where it creates queues when you look at the full journey of value.
  15. How to know if AI pays off: What to measure to know whether AI really improves anything, and how to turn it into continuous improvement.
  16. More than one team: the minimum you need to add: When you need to coordinate several teams and the minimum to add, with AI built into the coordination.
  17. Don't buy licences yet: How to manage AI adoption in a company with Scrum: a use case backlog, pilots with a stop criterion, and the move to production.
  18. Compliance does not have to slow you down: How to put the EU AI Act, the GDPR and human oversight into the backlog and the Definition of Done, instead of into a committee at the end.
  19. What does not change: Ten common mistakes with their antidote, a thirty-day plan and where to go next.

Chapters