Scrum was designed around a specific assumption: producing a testable increment takes real time, so batching decisions into sprints makes sense. Sprint Planning, Daily Scrum, Review, Retro. All of it was calibrated to how long it used to take a team to build something worth inspecting.
AI has compressed that side of the equation. A developer working with AI assistance can produce a working draft of a feature in a day, sometimes hours. The two-week box that used to match the pace of building no longer matches it. The rhythm is still there. The reason for the rhythm is not.
The team itself has changed
Previously, the development team meant designers, developers, and testers sitting together, each bringing judgment shaped by experience. Now some of those roles are agents.
That changes the ceremonies in ways Scrum has not addressed. What does a retro look like when half the “team” has no memory of the sprint unless someone feeds it back in? Does an AI agent join the Daily Scrum? If it does, what is it actually reporting: status, or a summary generated from logs it was told to summarize?
These are not edge cases anymore. I am seeing teams quietly work around this, running standups that are really just human check-ins about how the AI-assisted work is going, while the ceremony itself keeps pretending nothing has changed.
Scrum never solved the real problem
Here is the part I think gets missed. Building the right thing was always a human problem. Scrum did not solve it. It never claimed to fully solve it either, it just narrowed priorities through the Sprint Goal, giving the team one thing to rally around instead of ten competing ones. That was useful when the constraint was output: every option the team considered had a real build cost, so narrowing early was the responsible move.
That constraint is gone. Generating three competing approaches with AI is nearly free. The expensive part is no longer producing options, it is evaluating them well. Locking a team into a single Sprint Goal before that evaluation has happened is not protecting focus anymore. It is forcing a decision earlier than the evidence supports.
The problem Scrum was built to manage, deciding what is worth building, still exists. AI did not remove it. If anything, it made the problem sharper, because now teams can build the wrong thing faster than ever. Scrum does not have an answer for that. It never really did. It just made the symptom less visible by slowing everyone down to a pace where bad decisions took longer to compound.
Where this leaves teams
I am not arguing for chaos instead of structure. Human judgment, checking work against reality, catching what a model got subtly wrong, still needs a place to happen. But the fixed two-week container is the wrong place to protect that judgment when the team can finish something worth inspecting on day two and then has nothing to do with that information until the Review.
I have written before about why flow-based cadence tends to fit AI-accelerated teams better than fixed sprints, and about what happens when shipping outruns a team’s ability to decide what is actually worth shipping. Both of those posts are relevant here: Why Kanban Builds a Stronger Agile Mindset and AI Made Shipping Fast. It Made Deciding Slower.
Scrum is not obsolete. Teams with hardware dependencies, regulated releases, or heavy cross-team coordination still get real value from a fixed cadence. But for teams building with AI agents doing real work inside the sprint, the ceremonies need to be rebuilt around a different assumption than the one they were written for.
What would a Daily Scrum actually look like if half your team was an agent with no memory of yesterday?
Tahir Shahzad
Tahir Shahzad is a Product Manager, Product Owner, and technology consultant with over a decade of experience helping startups and organizations build products people actually use. If you are working on a product problem worth solving, reach him at tahirshahzad.com/contact/
