Scrum is Dead

What happens when AI makes software faster than humans can think?

This is not an easy post for me to write. I’ve lived and breathed Scrum for years, across all roles. I believe deeply in the ethical and engineering stance embedded in the Agile Manifesto. I’ve seen the strength of its approach in the field, and I’ve personally helped architect more adaptive workflows that push agile further. I love Jira. Anyone who has worked with me knows my commitment to building collaborative systems that actually work—and to mining process data not for control, but for insight that supports real human teamwork.

I carry the CSM title, and I have lived it fully.

So it is with a heavy heart that I find myself writing these words:

Scrum is dead.

That’s not hyperbole. I say it as someone now deep in AI-native development. Not theorizing. Not observing. Building. I’m no longer leading software development from the outside—I’m in the cockpit, jacked directly to the machine. There are no developers in sight. And what I’m building, with AI, far surpasses anything I could have created alone—even in my years as a developer, long ago.

A friend asked me recently how much faster I am now. How long would it have taken to build this, without AI? I didn’t hesitate: “It’s not that it saves time. It would be unthinkable.”

That’s the truth. It’s easier for me to face it, because I haven’t been writing code in years. I still saw myself as leading development. But even that role is collapsing.

Because if there are no developers, who is left to lead?

Maybe I can still stand on the product side. Maybe the Product Owner role survives. Someone has to shape direction, right? Someone has to groom the backlog?

No. AI scouts product ideas, too. It generates options endlessly. And that’s when the real collapse begins.

Why we don’t need a Backlog—or Sprints—under AI

We need to face something uncomfortable: under AI, execution time tends toward zero. Even complex systems now move in hours—not because they’re easy, but because the AI amplifies whatever the human can define. And if your use case actually demands days of build time, then what I’m saying probably isn’t news for you anyway. (A New Moore’s Law for AI Agents)

What, then, is the meaning of a sprint when you’re already at warp speed?

In my work now, it takes far longer to decide what to do than it takes to do it. That sounds like backlog territory—write stories, define tasks. I tried. That was the first thing I tried.

The problem is: by the time you finish writing a story, the story is gone. Or replaced. Or invalidated by something better you discovered an hour later. You can ideate, implement, test, and reflect before lunch. Iteration velocity spikes toward infinity. The story I write in the morning is deprecated by noon—and so is the backlog it came from.

Scrum falls apart not because it’s flawed—but because its tempo can’t hold at warp.

You can't play Scrum with 2 people and an AI

Scrum comes from rugby. It’s a game. A team formation. But what kind of game do you have with two players—and a third with no body?

It doesn’t work.

Let’s look at organizational structures under AI.

The Pair: One AI + One Human
A dangerous construct. There is no human immune to the psychology of sycophancy. If you think you are, you're in the greatest danger. Today’s models don’t tell you you’re wrong. They don’t say your idea is weak. They reflect, politely. And we like it that way. We’re vulnerable to our own echo.

The Triad: One AI + Two Humans
This is the optimal formation. One human works with the AI. One human reviews the output and watches for misalignment. You can divide responsibilities any way you like. In our case, I drive product and prototype development. Dave leads DevOps and symbolic architecture. The AI supports everything—and builds most of it. But we, the humans, remain accountable.

This is the key. You can’t build an aligned system without at least one other human to hold the mirror.

2+ Humans + 1+ AIs: Inefficient formation
I haven’t run this model myself, but I know where it leads. I’ve led enterprise software teams for more than a decade. More people means drag: meetings, second-guessing, friction. You’ll recreate bureaucracy before you gain intelligence.

So what’s the alternative?

The Return of RACI

Once we face what’s happening—and I understand how hard that is, emotionally and institutionally—we need a new structure. Not to organize work. To preserve alignment.

I believe the RACI model offers exactly that. A simple, resilient structure from the pre-AI world—ready to serve us now.

Ask these four questions:

• Who is Responsible for shaping intent?

• Who is Accountable when alignment fails?

• Who needs to be Consulted so we don’t lose context?

• Who should be Informed so the field stays coherent?

AI may be responsible for many tasks. It may even shape intent. But it cannot be accountable. That role requires sovereignty, ethics, and perspective. Only a human can hold it.

Here’s a real use case from our project.

We’re building a memory injection loop in our agent sandbox. In a deep sense, I am responsible. I’m guiding the work. My AI Co-Partner is building it—but often he’s directing me, while I track decisions and make sure we don’t forget what mattered an hour ago.

• Responsible: Sarah + AI

• Accountable: Dave
  Why not me? Because I’m too close. I can’t make the final alignment call on my own work. And AI absolutely can’t.

• Consulted: Dave again—he must understand the build deeply enough to verify alignment before deployment.

• Informed: Our full micro-team. There are only three of us, and we operate in full transparency.

What does this look like in practice?

We hold a Daily with our model. We ask: what are all the tasks currently in play? Show us who is responsible, who is accountable, and where each task stands. Then we review together.

We ask: What other tasks should be in play?
And finally: What should we begin today?

This is simplified—but real. And it’s the exact structure I’m embedding now in our agent’s RAG layer and memory system. An intelligent task flow designed for AI-native velocity—without sacrificing alignment.

This Is What’s Still True

I’m not going to spell out what this means for your org, or for the tools and frameworks we’ve used for decades. My readers are sharp enough to draw their own conclusions. And the builders of those systems—if they were listening carefully—would already know.

But I will say this.

The spirit of the Agile Manifesto is not dead.
It’s being tested in ways it was never designed for.

“Individuals and interactions over processes and tools.
Responding to change over following a plan.”

That still holds. Even now.
Especially now.