Updated
What does Anthropic's AI-native SDLC playbook mean for Scrum teams?
Anthropic's playbook says code is no longer the bottleneck, because planning, review, testing and deployment still run at human speed. For a Scrum team that puts the pressure on the Definition of Done, the Sprint Review and Product Owner decisions. My take: Scrum doesn't change. Weak inspection and unclear accountability show up faster.
What the Scrum Guide says
The Scrum Guide does not mention AI. It does say what makes work count, how the Sprint Review works, and whose decisions the organization must respect:
"Work cannot be considered part of an Increment unless it meets the Definition of Done."
"The Sprint Review is a working session and the Scrum Team should avoid limiting it to a presentation."
"For Product Owners to succeed, the entire organization must respect their decisions." Read the Definition of Done section of the Interactive Scrum Guide.
Go deeper
I covered this in two videos. Part 1, Why AI Won’t Speed Up Your Scrum Team (The 1999 Trap), is about where the extra code piles up; its resource page has the transcript. Part 2, Code Got Cheap. Accountability Didn’t., walks a fictional Scrum Team through the playbook; its resource page has the transcript. For the short answers, read Will AI make my Scrum team deliver faster? and How should the Definition of Done change when developers use AI?
What the playbook says
Anthropic’s playbook, published August 21, 2026, argues that AI now writes code faster than the rest of the delivery process can absorb it. The steps around build, mainly planning, review and testing, and deployment, “still run at human speed.”
Its answer is to rebuild those steps around the AI while people stay in charge. In its words, “Humans remain accountable for every decision that requires judgment.” The playbook never mentions Scrum, so the translation below is mine.
What it means for a Scrum team
- Definition of Done. Review and testing are where AI-written work waits, so that is where Done gets decided. My take: write the Definition of Done so it can be checked while the work happens, and name the person who approves.
- Sprint Review. More output makes it easier to show a lot and inspect little. My take: show stakeholders a working Increment they can use, not a pile of generated drafts.
- Sprint Planning. Planning is one of the human-speed steps. My take: plan to what review and testing can finish this Sprint, not to how fast the team can generate code.
- Accountability. The Product Owner still decides what is worth building, and the Developers still answer for what they ship. An AI tool is accountable for neither.
What I see in practice
I made the same point in Part 1: “AI tools don’t remove delivery bottlenecks. They move them.” In Why AI Won’t Speed Up Your Scrum Team (The 1999 Trap) I wrote that the extra code lands in “peer review, the test suite and Sprint Backlog refinement, and those still run at human speed.”
When I read the playbook, my first reaction was “Waterfall, in markdown.” Then I read it again. In Code Got Cheap. Accountability Didn’t. I called it a loop, not a staircase, and said the Definition of Done is “enforced while the work happens, not argued about at Sprint Review.” On who answers for the result, I put it this way in Part 1: “Accountability remains 100% human.”
The number to watch is in Synthetic Speed: “Stop measuring how fast your teams generate drafts, and start measuring how quickly working software reaches real users.”
Common mistakes
- Reading “code is no longer the bottleneck” as “the team can take on more work.”
- Speeding up the build and leaving review, testing and deployment as they were.
- Adopting the playbook’s files and steps as a checklist, without inspecting whether they help your team.
- Letting the AI approve its own work, or treating green tests as proof when the AI wrote them.
- Treating AI accountability as a tooling question. It is a question about people.
What to do this week
- Look at your last Sprint. Find where work waited for review, testing or approval, and write down for how long.
- Read your Definition of Done with the Developers. Ask which items could be met without any person having understood the change.
- Pick one human-speed step to improve in your next Sprint, and inspect the result in the Retrospective.
FAQ
Does Anthropic's playbook mention Scrum?
No. It describes six stages of a software delivery lifecycle (plan, design, build, test, deploy and maintain) and does not use the words Scrum, Sprint or Definition of Done. The translation on this page is mine. For the Definition of Done in particular, see How should the Definition of Done change when developers use AI?
Who approves the work in the playbook?
A person does. The playbook says the agent that wrote the code cannot approve it, and that approval comes from a human. It also has the product owner decide whether the intent and spec move on to build. My take: that is the same split Scrum already has between the people who do the work and the people accountable for what it is worth.
Should a Scrum team adopt the playbook's workflow as written?
My take: not as a checklist. The playbook is written around using Claude in engineering organizations. In Code Got Cheap. Accountability Didn't. I said to treat it like anything else in Scrum: try it, inspect it and adapt it, with the whole team.