Is AI Becoming the New Snake Oil?

Every few years, someone declares Scrum dead. Now AI has apparently killed it again.

The argument goes something like this: AI can generate software dramatically faster, developers can accomplish far more individually, and agents can increasingly plan and execute work themselves. Therefore, the meetings, refinement, planning, estimation, and other machinery that organizations have built around Scrum no longer make sense. There’s some truth in that argument, but there’s also a very familiar trap.

Scrum Isn’t the Point

I’ve never been particularly interested in defending Scrum for the sake of Scrum. It’s a tool. Where Scrum fits the nature of the work, is practiced well, and respects the people and teams doing that work, it can still be useful. Where it has become rigid ceremonies, ticket factories, endless refinement, story point debates, and process for process’s sake, change it or get rid of it.

That was true before generative AI arrived. The annual “Scrum is dead” debate tends to distract us from a much more important question: What problem are we actually trying to solve? Replacing one methodology with another doesn’t necessarily improve the system, and neither does replacing a methodology with AI.

We’ve Seen This Movie Before

Many Agile transformations promised enormous improvements in productivity. Organizations reorganized teams, trained thousands of people, adopted Scrum, installed Jira, hired coaches, and measured velocity. Yet many still struggled to achieve the business outcomes they expected.

One reason is surprisingly simple: they improved one part of a much larger system. A development team might deliver software faster, only to have that work sit waiting for integration, security review, testing, deployment, legal approval, operations, or a business decision. The team became faster, but the system didn’t. Sometimes making one part dramatically faster actually made the overall problem worse by creating a larger queue at the next constraint. AI may be taking us down exactly the same road.

What Happens When Code Becomes Cheap?

Suppose AI really does allow developers to produce several times more software. That could be an extraordinary capability, but can the rest of the organization absorb it? Can testing validate several times more changes? Can security review them? Can operations safely deploy and support them? Can customers absorb several times more features? Can product teams determine whether those features should have been built in the first place?

Perhaps the most important question is whether the organization can learn several times faster. If it can’t, then increasing the rate at which we produce code hasn’t necessarily increased the rate at which we create value. We’ve simply moved the bottleneck, and perhaps flooded the rest of the system with more work. This is basic systems thinking, yet it has been missing from far too many Agile, product, and organizational improvement efforts.

Local Optimization Is Still Local Optimization

Organizations love local optimization because it’s relatively easy to see and measure. Developer productivity increased. Cycle time dropped. AI generated most of the code. Teams shipped twice as many features. Those numbers can look fantastic on a dashboard, but customers don’t buy developer productivity, story points, lines of code, AI prompts, or sprint velocity. They experience the output of the whole system.

That distinction becomes even more important in the AI era. If AI makes coding almost instantaneous while everything around coding remains unchanged, software development hasn’t become instantaneous. One activity within a much larger value stream has become faster. The constraint simply moves somewhere else.

AI Should Change How We Work

None of this means organizations should preserve existing Agile practices because “that’s how Scrum works.” Quite the opposite. If AI reduces a feedback loop from weeks to hours, waiting two weeks to inspect what happened makes little sense. If an agent can perform analysis that previously required a meeting, perhaps that meeting should disappear. If detailed backlog refinement no longer adds value, stop doing it.

We should shorten feedback loops, reduce unnecessary coordination, push decisions closer to the people doing the work, and give teams greater autonomy to experiment, learn, and adapt. Ironically, those ideas aren’t particularly anti Agile. They’re pretty close to what Agile was supposed to encourage in the first place.

The Snake Oil Problem

My concern isn’t AI. I use AI, and I think it will substantially change how knowledge work and software development happen. The snake oil is the idea that adopting AI somehow removes the need to understand and improve the system around it.

We’ve heard versions of this promise before. Adopt Agile and everything will move faster. Install Scrum and teams will become productive. Move to the cloud and the organization will become digital. Now we’re hearing that giving everyone AI agents will turn everyone into a 10x worker.

Perhaps AI really can make an individual ten times more productive at certain activities. But ten times the output from one part of a constrained system doesn’t produce ten times the value. Sometimes it just produces a bigger queue.

Stop Asking Whether Scrum Is Dead

Scrum will work in some environments and poorly in others. AI will make some Scrum practices unnecessary, compress others, and perhaps make rapid inspection and adaptation even more valuable. That’s exactly what we should expect as the nature of the work changes.

The more useful conversation is about whether we’re improving the whole system through which ideas become customer value. So rather than asking whether AI has finally killed Scrum, I think there’s a better question: If AI makes one part of our organization dramatically faster, what becomes the constraint next?

Find it, improve it, and then look at the system again. Because the goal was never to make developers faster, implement Scrum correctly, or deploy more AI. The goal is to make the whole system better.

Explore this topic

Related articles

Is AI Making Organizations More Effective, or Just Busier? Consider whether AI is improving overall effectiveness or simply increasing activity and review work.

Agile at Scale: How to Avoid Missing the Point Consider why expanding a framework can miss the values and collaboration that make Agile useful.

Hear the ACN conversation

ACN Episode 112: Is AI Redefining Leadership? Hear the community explore how AI changes coordination, leadership work, and the responsibilities people still carry.