I’ve spent the last 25 years trying to answer one question: how do you actually get a large, complex organization to deliver software predictably? The question hasn’t changed. The technology has changed. The buzzwords have changed. The funding model has changed. But the underlying problem has been the same one the whole time, and right now it’s wearing a new costume called AI.
So when CIOs ask me what they should be telling their board about AI, I tell them the same thing: don’t tell them about AI. Tell them about the preconditions.
The Pattern I Keep Watching Repeat Itself
If you’ve been following our work for a while, you’ve heard me say that agile fails when the structural preconditions for it to work aren’t in place. We’ve been saying that for almost 20 years. It was true when the new thing was Scrum. It was true when the new thing was DevOps. It was true when the new thing was cloud migration. It’s true again now that the new thing is AI.
The pattern goes like this. A new technology emerges that promises a step-change in business outcomes. Leadership gets excited. Pilots get funded. The pilots produce a couple of demos that look great in a boardroom. The demos don’t scale into the actual business. Six months later the conversation turns to how much the team is learning. 18 months later the initiative quietly gets de-funded and replaced with the next new thing.
I’ve watched that arc play out with every major tech wave in the last 25 years. It’s playing out right now with AI. It will play out next with whatever comes after AI. The pattern is a structure problem.
Why Most AI Pilots Aren’t Scaling
A partner of ours said something to me about a year and a half ago that stuck: the move to AI is going to force digital transformation to become real. He was right. There was a lot of room in the cloud-migration wave for a CIO to spin up some containers, claim victory, and tell the board they were digital. AI doesn’t grant that kind of room. AI either solves a real business problem using your actual data and workflows, or it doesn’t.
Most of them don’t, and the reason is structural. Your AI pilot is sitting in a system that wasn’t designed for it. The data it needs is locked inside a system of record from 30 years ago that runs on a different operating assumption. The workflow it’s supposed to accelerate runs through six teams that don’t share a backlog. The output it produces has to be trusted by an executive whose dashboard is wired into a different governance system entirely. The pilot demo works because the demo conditions are controlled. The production reality isn’t, and the production reality is what your board cares about.
I’ve been saying for a long time that the unit of agile delivery is a complete cross-functional team with a clear backlog producing a working tested increment of the product on a regular cadence, with no dependencies they can’t manage themselves. Teams, backlogs, working tested software. That’s it. That’s the whole frame. Every word in that sentence is doing real work, and every word of it applies just as much to AI as to anything else we’ve ever tried to ship.
If your AI team is six engineers spread across three reporting lines, dependent on a data team in another building, releasing into a workflow they don’t own, against KPIs nobody has agreed on, the outcome is predictable. Your AI pilot will behave exactly the way every failed agile pilot has behaved for 20 years. In any useful sense, it’s a structure problem wearing AI clothes.
What I’d Actually Tell Your Board
Here’s what your board needs to hear: we’re going to be predictable.
If I were the CIO, here’s how I’d frame the board conversation. We are going to deliver AI-enabled business outcomes every 90 days, against a small number of capabilities at a time, and we’re going to deliver them predictably. We’re not going to commit to a portfolio of 15 AI use cases on a slide somewhere and then come back in 18 months apologizing for what didn’t ship. We’re going to commit to one capability that we can actually land, ship it, and earn the right to fund the next one with the value we just shipped.
That is a delivery commitment the board can hold us to. It’s the same predictability promise I’ve been trying to get technology organizations to make for 20 years. AI doesn’t change the standard. If anything, AI raises it, because the cost of AI tooling and AI talent makes a slow, unpredictable AI program more financially painful than a slow, unpredictable agile transformation ever was.
The reason this lands at the CIO level is that you’re the one who has to translate the structural reality into board language. Engineering can talk to engineering about modular architecture and complete cross-functional teams. The CFO can talk to the CFO about CapEx and OpEx. You sit in between. You’re the one who has to take “we need to break the dependencies between our systems before AI can scale” and turn it into “we will deliver business outcome X by date Y at confidence level Z.”
That’s a structural conversation, not a tooling conversation. And it’s the conversation our industry has been bad at having for a long time. We tend to want to tell the board about the new thing we bought. The harder, more honest CIO move is to tell the board about the old thing we’re redesigning so the new thing can work, and to commit to a delivery cadence the board can verify against the calendar.
Where to Start
If you take one thing from this post, take this. The structural work and the AI delivery aren’t sequential. They happen inside the same 90 days, against the same capability, by the same team.
Pick one business capability where AI would materially matter. Not five. One. Make it the one you’d most want to win. Then ask three questions about it.
One. Is there a single team that owns this business capability end-to-end, with no external dependencies that they can’t resolve themselves? If not, your first move inside the 90 days is forming that team.
Two. Is the data this AI needs encapsulated in a service that team can call, or is it sitting inside a legacy monolith that has to be touched every time something changes? If the latter, part of the 90 days is extracting and encapsulating just enough of that data for this one capability to work cleanly. Not all of your data. The slice this capability needs.
Three. Does the workflow this AI is meant to accelerate have a clear, measurable business outcome that an executive sponsor will hold their team to? If not, the 90 days include defining that outcome, in operational terms, with the sponsor signed up to it.
By the end of the 90 days, three things are true. The structural preconditions for this capability are in place. The AI is deployed against the capability in production. There is a measurable business outcome that the team and the sponsor agree was hit. That’s your first delivery. That’s what you take to the board at the end of Q1.
Then the next 90 days add the next capability. By the end of year one, you’ve got four capabilities running on the new structural pattern, each with AI compounding on top of it, each with a clean business outcome the board can point to. Most CIOs don’t see this coming. The structural work you did for capability one pays forward into capabilities two and three. The team gets faster. The data layer gets cleaner. The governance model gets more practiced. The cadence becomes the operating rhythm.
That’s a very different story than “we’re not delivering AI this year.” It’s “we’re delivering AI quarterly, against the capabilities that matter, on a cadence you can take to the audit committee.” Same underlying preconditions argument. Very different message in the boardroom.
Most of the CIOs I talk to are already deep into an AI program that doesn’t look like that. The move isn’t to throw it out. The move is to pick the one capability inside it that’s most worth saving, restructure the work around that one, and let the unsalvageable pieces wind down naturally as funding cycles end. Don’t take a year off from AI to prepare. Take 90 days to do one capability right, then prove the pattern repeats.
The firms that figure this out in the next two years won’t be the firms that bought the most AI tools. They’ll be the firms that learned to deliver AI predictably against a small number of capabilities at a time. The rest will spend year three explaining to their boards why the AI bill keeps going up and the business outcomes haven’t moved.
That’s the story. Tell that one.
