Before I jump in, let me clarify something: Agile methodology and agility are not the same thing. Agile is a set of practices, a generation’s best answer to a specific set of constraints. Agility is something older and more fundamental: the ability to sense what’s happening, respond intelligently, and keep adapting. We never evolve past that. It is as relevant today as it was the day someone first wrote it on a whiteboard, regardless of whatever comes next.
Which brings us to AI, and the question every team is quietly wrestling with: how do we stay agile in a world where the tools just changed dramatically?
The answer, as it has always been, is to find your bottleneck.
The Bottleneck Is Always the Story
Every methodology, every process improvement, every framework ever invented has ultimately been an attempt to answer the same question: where are we losing time, energy, and clarity, and how can we improve?
Agile’s early practices were designed around a very specific bottleneck: human execution pace. Developers could only write and test so much in a given window. Work expanded to fill the time available. So we built practices that matched planning to that pace: small increments, just-in-time decisions, small teams to optimize communication, two-week cycles that kept teams focused and fed without overwhelming them.
Those practices worked because they addressed the real constraint. But for teams leveraging AI, that constraint has now moved.
AI tools can execute in hours what once took days. They don’t get pulled into meetings, don’t lose context after a long weekend, and don’t have three competing priorities. The execution bottleneck that Agile was largely designed to address has disappeared.
So where did it go?
Finding the New Bottleneck
This is where visualizing your work becomes essential. We recently wrote about how important it is to make your workflows visible, and that principle hasn’t changed. If anything, it becomes more important. AI is part of your workflow now, but that doesn’t make it invisible. The same questions we’ve always asked about value streams still apply.
Where are things slowing down? Where are tasks getting dropped? Where does clarity break down?
For most AI-assisted teams, the new bottlenecks tend to cluster in a few places:
- The quality of human direction. AI builds what you ask it to build, with impressive speed and very little pushback. That means the clarity of your prompts, requirements, and direction has become a primary constraint. Vague input used to slow a human developer down while they asked clarifying questions. Vague input to an AI produces a fast, confident, and potentially very wrong result.
- Review and feedback cycles. If AI can generate a week’s worth of work in a day, but your review process still operates on a human weekly cadence, you have a new bottleneck. Work is piling up behind the checkpoint.
- Decision-making authority. Who can approve the next chunk of work? Who decides when a feature is ready to test? In fast-moving AI-assisted teams, unclear decision rights create invisible queues that nobody is tracking.
- User feedback loops. AI is extraordinarily good at building things. It cannot tell you how real humans will respond to what it built. The faster you can execute, the faster you can generate a mountain of beautifully-built wrong turns. We need to see how our real human users interact with what we’ve built.
Big Up-Front Planning: One Example of How the Math Has Changed
The bottleneck shift shows up in interesting ways when you look at specific Agile practices. Take the long-standing guidance against big up-front planning.
That guidance made sense when execution was the constraint. Spending two weeks planning work that would take six weeks to build was wasteful because a lot would change in those six weeks. Better to plan a little, build a little, learn a little.
But here’s what AI actually changes about that cycle: the benefit isn’t just that the work goes faster. It’s that the whole feedback loop goes faster: plan, execute, get feedback, adjust, try again. What used to take weeks can now take days. Military strategists call this the OODA loop (Observe, Orient, Decide, Act) and the competitive advantage has always gone to whoever can cycle through it fastest. AI compresses that loop dramatically, and that’s where the real opportunity lives.
Which makes it all the more tempting to do the wrong thing with it. AI engines have almost limitless work capacity, and it’s easy to get lured into treating that as the point. Feed them more. Build more. Move faster. But as Jeff Goldblum’s character warns in Jurassic Park, “Your scientists were so preoccupied with whether or not they could, they didn’t stop to think if they should.” Volume of output was never the bottleneck. The ability to learn is the bottleneck.
So the answer isn’t more planning or less planning. It’s appropriate planning, calibrated to support faster learning rather than just faster building, with feedback loops built in before the next cycle starts.
AI Changed the Process. It Didn’t Change the Mission.
Your goals are the same as they were before AI arrived. Build things people want. Reduce waste. Get feedback quickly. Deliver value. None of that gets outsourced to a language model.
What AI changed is the process, and that’s worth taking seriously. A new tool this powerful deserves a genuine examination of your workflows. Where are you losing time? Where is clarity breaking down? Where are experiments getting run? What does your feedback loop from users actually look like right now, and is it fast enough to keep pace with what your team can build?
These are the questions agility has always asked. The practices we use to answer them may need to evolve. The asking never stops.