From time to time, somebody declares Scrum lifeless. Normally, it coincides with the arrival of one thing shiny, a brand new methodology, a brand new instrument, a brand new motive to consider that this time, complexity has lastly been solved. This time, the obituary writers have AI on their facet, and I’ll grant them this a lot: they’ve a degree, simply not the one they suppose they’re making.
I run Scrum.org, so take my bias as learn. However what I’m about to argue isn’t a protection of an establishment; it’s a warning about what occurs to improvement groups that abandon construction exactly after they want it most.
Right here’s what’s truly occurring. AI is making particular person builders dramatically sooner. Code will get generated in seconds. The boilerplate that took a day is finished earlier than the espresso is poured. Should you measure a single developer’s output on a structured, well-defined activity, AI can ship an actual enchancment. That a part of the story is true, and anybody who tells you in any other case is promoting one thing.
However right here’s what the productiveness fanatics aren’t telling you. Quicker people don’t mechanically produce higher groups. In actual fact, the proof suggests they usually produce worse ones. McKinsey’s 2025 State of AI report surveyed almost 2,000 organizations throughout 105 international locations and located that whereas 88% now use AI commonly in not less than one operate, 94% report not seeing important worth from these investments. Take into consideration that for a second. Close to-universal adoption, negligible enterprise-level influence. Particular person notion of productiveness, flat organizational actuality.
The velocity paradox
The impartial analysis on improvement groups particularly is much more putting. METR, a non-profit analysis institute, ran a randomized managed trial in 2025 with skilled open-source builders engaged on their very own codebases. Builders utilizing AI instruments took 19% longer to finish duties than those that didn’t. The kicker: those self same builders estimated afterward that AI had made them 20% sooner. The hole between what they believed and what the information confirmed isn’t just a curiosity; it’s a warning sign about how completely AI can distort our capability to evaluate our personal efficiency.
In the meantime, a longitudinal evaluation of 211 million traces of code by GitClear discovered that as AI coding adoption rose, refactoring exercise collapsed from 25% of code modifications in 2021 to beneath 10% in 2024, whereas code duplication elevated eightfold in a single 12 months. Extra code, sooner, with much less high quality, much less reuse, and a mounting upkeep burden that hasn’t proven up in anybody’s velocity metrics but.
I name this the velocity paradox. AI accelerates the person and, in doing so, can quietly destroy the group. Quicker people diverging on totally different duties, with totally different assumptions, producing extra output with much less shared understanding, create not a productiveness engine however a coordination disaster. The bottleneck doesn’t disappear. It strikes upstream, from execution into alignment, judgment, and governance. Which is exactly the place Scrum lives.
Scrum was designed for precisely this sort of downside. Not for predictable work the place necessities are identified and the trail is evident, any first rate venture plan handles that. Scrum was designed for complexity, for conditions the place the reply can’t be identified prematurely and the place the one smart response is to make small bets, be taught quick, and alter course. Dash Targets drive a group to agree on a single helpful consequence quite than a set of duties. The Dash Evaluate brings actual proof of worth to stakeholders quite than a quantity of shipped code. The Definition of Carried out prevents AI-generated output from silently accumulating debt. None of that issues much less when AI is within the room. It issues extra.
Functionality debt
There’s a subtler downside beneath the velocity paradox that I need to title, because it hardly ever comes up in AI adoption conversations. That is referred to as functionality debt. When AI takes over the execution duties, writing the operate, drafting the check, and producing the documentation, these are the identical duties that used to construct junior builders’ judgment. Anthropic’s personal researchers, Judy Shen and Alex Tamkin, ran a randomized managed trial in 2026 to review how builders realized a brand new programming library with and with out AI help. Members who used AI scored 17% decrease on comprehension assessments than those that coded by hand, with out ending meaningfully sooner. The most important hole appeared particularly in debugging capability, which is exactly the talent you want most when your job is to supervise and consider AI-generated output. We’re, in different phrases, utilizing AI to skip the educational that makes people able to supervising AI. That may be a compounding downside, not a one-time value.
This connects to the truth that most organizations misunderstand AI fluency. The error is treating it as an individual-capability downside: give everybody entry to instruments, run a couple of coaching periods, and watch productiveness climb. However particular person fluency is just not group functionality. There are significant ranges at which practitioners have interaction with AI. On the foundational stage, the first concern is protected, intentional use with out passing on hallucinations. On the practitioner stage, the place most builders sit at this time, fluency improves particular person output however introduces a refined lure: the extra effort you spend money on crafting a immediate, the extra possible you might be to belief the end result with out adequately scrutinizing it. Higher prompting can paradoxically decrease your analysis guard. The antidote to that lure isn’t higher instruments. It’s the form of shared evaluate, collective accountability, and structured inspection {that a} well-run Scrum Workforce builds as a matter of apply. The third and most respected stage of fluency is about designing the group’s AI workflow, not simply your personal, and it’s inherently a team-discipline downside.
The groups getting probably the most real worth from AI usually are not those that deserted construction. They’re probably the most disciplined ones, the place Dash Targets give AI a course to work towards, the place a shared Definition of Carried out prevents technical debt from quietly accumulating, and the place the Retrospective institutionalizes AI studying on the group stage quite than leaving it to particular person experimentation.
So no, Scrum isn’t lifeless. We’re on the exact second the place empiricism, shared targets, collective accountability, and structured inspection develop into the distinction between AI making your group genuinely higher and AI making your group sooner at producing the fallacious issues. These usually are not the identical consequence, and the hole between them will outline which improvement organizations look again on this second because the 12 months they received it proper, and which of them are nonetheless untangling the implications.
