As AI coding tools become more common in software development, many organizations are questioning whether their agile practices still hold up.While this concern is legitimate, it is also misdiagnosed. Many teams were already running a hollowed-out version of agile principles before AI arrived, which is why AI is now accelerating both the good and the bad.This article examines what is genuinely changing about agile practice in an AI-assisted environment, what risks teams need to manage, and what organizations should prioritize to drive efficiency and stay competitive.
When the Rituals Outlive the Reasoning
Agile methodology had several phases, but it began as a practitioner movement. In the early 2000s, a group of software practitioners published a one-page document, the Agile Manifesto, built around four value statements and twelve principles. Small teams pushed back against slow, process-heavy development by committing to short iterations, frequent feedback, and working closely with the people they were building for.The core values were straightforward: working software over comprehensive documentation, customer collaboration over contract negotiation, responding to change over following a plan, and individuals and interactions over processes and tools. In practice, this meant shipping in short cycles, checking in often, and treating feedback as the primary source of truth about whether the work was going in the right direction.Back then, the approach worked because the principles were applied with intent. Things started changing as soon as enterprises scaled it up. Consulting firms and certification bodies turned agile into something you could buy, and the focus shifted from thinking differently to complying with a framework.Retrospectives would produce the same action items quarter after quarter. Sprint reviews turned into demos that stakeholders rarely attended. Lance Dacy, a certified Scrum trainer, describes this pattern as “agile theater”: going through the motions without internalizing the principles.According to Forrester’s 2025 State of Agile Development report, 95% of professionals globally affirm that agile principles remain critical to their work, yet only 7% of organizations report achieving full proficiency. These findings indicate that most organizations got good at performing agile, but very few got good at practicing it. The advent of AI has only made it harder to ignore these gaps.
What AI Actually Changes (And What It Does Not)
AI coding tools did not change what good software development looks like. They changed who (or what) does the work.In the traditional agile model, a developer wrote the code while stakeholders evaluated working software at sprint reviews, and developers peer-reviewed each other’s code internally. These were two distinct feedback loops running in parallel.Today, a coding agent often drives a large portion of the implementation. The developer shifts from author to something closer to an editor or director:
Describing the intent
Reviewing the output
Directing the next step
That shift sounds straightforward, but it carries a less obvious consequence: accountability changes shape. When AI writes the code, the developer owns the decision to accept it. That is a different cognitive task than writing the code in the first place. It requires judgment about correctness, fit, and intent, and often across a large volume of output.The feedback loop is nested differently, but it is still there. And because an AI agent can generate a substantial volume of code in minutes, the discipline around how that work is sized and reviewed becomes more important.If a large AI-generated output lands as a single pull request, the review process becomes unmanageable. Reviewers skim, focus on a fraction of the changes, and approve the rest on faith. In these circumstances, and due to the limitations of human attention, understanding is bound to break down. A unit of work should be scoped to what a reviewer can hold in their head at one time: substantial enough to be meaningful, contained enough to be understood. Keep in mind that this applies specifically to pull request sizing, not to sprint scope.Pacing works the same way. The Agile Manifesto calls for sustainable development: “sponsors, developers, and users should be able to maintain a constant pace indefinitely.” AI makes it easy to unknowingly run past that ceiling. The output feels manageable, until it does not. Evan Phoenix, author at Miren, describes his own practical ceiling at roughly three AI agents running in parallel before decision quality starts to degrade. That threshold is likely to vary by individual, but the underlying principle applies broadly. More agents do not mean more productivity if the person directing them cannot keep up.
The Risks of Getting AI Agile Wrong
Applying AI on top of processes that were already not working is the most significant risk organizations face right now. Each of the changes described above creates a corresponding failure mode when the foundation is weak.Shallow reviews get shallower. If code review was already a rubber-stamp exercise before AI, AI-generated output will make it worse. When AI produces the work and humans approve it without meaningful engagement, shared understanding erodes. Teams lose track of what was built and why. Review is a sync point allowing teams to stay aligned on what they are building together. Strip that out, and the codebase becomes something the team navigates by memory and assumption rather than shared knowledge.Faster delivery does not fix broken decisions. If sprint decisions were already made before the sprint review, accelerating the delivery cycle will not change that. If strategic direction shifts quarterly while teams ship weekly, the effective feedback loop is still three months long regardless of how much AI compresses the technical work. Dacy identifies three practical questions worth asking:
Do feedback loops get shorter over time?
Do team events drive decisions, or confirm ones already made?
Does the team adapt based on evidence, or based on whoever is loudest in the room?
Human sustainability is a real constraint. AI tools create pressure, sometimes explicit, sometimes cultural, to fill every saved hour with more work. Better tools should produce better outcomes, not more hours of output. The cognitive load of directing AI work is real and new. Running at full capacity feels sustainable right up until it does not, and recovery takes longer than the time saved. That is why identifying a pace the team can maintain over weeks, not just a single sprint, matters more than ever.
What to Prioritize Going Forward
Organizations handling this transition well are applying agile principles more carefully, not abandoning them.Protect the feedback loop. Short iterations only create value if the organization acts on what it learns. If you cannot answer Dacy’s first question on feedback loops getting shorter, that is the place to start. Make sure strategic decisions move at the same speed as delivery, or the speed of delivery is irrelevant.Right-size the work. With AI in the loop, the temptation is to batch up large outputs and push them through quickly. Resist it. Keep pull requests scoped to what a reviewer can engage with meaningfully. If the change is too large to discuss coherently, it is too large to review well. Splitting work deliberately is not inefficiency, but simply how shared understanding survives.Make review a collaboration point, not a gate. Every change should receive human attention to maintain shared understanding of what was built and why. When that step becomes a formality, the team stops knowing what it owns.Set a sustainable pace deliberately. Ask what pace the team can sustain for a month, not just this sprint. If the work is done, the day is done. The pressure to fill saved hours with more output is real, but it leads to the kind of accumulated cognitive debt that does not show up until it is expensive to fix.Measure what matters. Velocity is not a proxy for agility. The right questions are: Are feedback loops getting shorter? Are decisions being made on evidence? Is the team adapting based on what it observes, or executing against plans that were set before the work started?
The Competitive Difference Is Execution, Not Tooling
AI does not separate high-performing teams from the rest. The gap was already there. What AI does is make it more visible, more quickly.The organizations that thrive are protecting the same things they always should have been: short feedback loops, genuine willingness to adapt, and leadership that rewards learning rather than compliance. As Dacy observes, the organizations people point to as having “moved past agile” are, on closer inspection, still doing autonomous teams, iterative delivery, and continuous customer feedback: the Agile Manifesto’s principles under a different name.Agile principles were designed for complex, fast-moving, uncertain environments. That description fits the current moment precisely. The work is to take a closer look at the foundations, apply them with intent, and stop mistaking the rituals for the reasoning behind them.The tools are faster. The principles are the same. The question is whether the team is actually using them.
