Around 2010, a lot of organizations discovered Scrum. The consultants arrived, the training happened, the ceremonies were scheduled, and within six months most of those companies were running daily standups and sprint reviews with great fidelity. Two years later, most of them had the same throughput they had always had. The Scrum board was meticulous. The velocity metric was being tracked and reported to leadership. Nothing shipped any faster. Nothing changed about who had to approve what before anything left the building.

I spent years inside those organizations. What I watched happen was not a failure of the tool. It was a failure to ask the harder question before the rollout started: what does the operating model need to look like for this to produce results? That question requires mapping actual decision authority, workflow structure, approval chains, and success metrics. It takes political will to answer because every answer involves someone's authority or someone's identity. Most organizations skipped it entirely. They deployed the tool and waited for the outcomes to follow.

In 2026, I am watching it again.

The Data Is More Familiar Than It Looks

According to Writer's enterprise research published this year, 79% of organizations are struggling to extract value from their AI investments, a double-digit increase from the prior year. A separate study found that 95% are seeing zero ROI from their GenAI deployments specifically. The investment levels are not trivial: more than half of companies are spending over $1 million annually on AI technology. The boards want progress. The C-suite is under real pressure. Seventy-three percent of CEOs report stress or anxiety about their company's AI strategy, and 64% say they fear losing their jobs if they fail to lead the transition well.

And 29% of employees are quietly sabotaging the rollout.

That last figure is the one that tells you what is actually happening on the ground. One in three workers is not resisting AI out of technophobia or generational stubbornness. They are watching their organization automate a process that does not work, add AI-generated output to a review cycle that still takes three weeks, and mandate tool adoption in a workflow where the tool solves no actual problem the worker has. When you have watched the same play run before, you stop pretending this iteration will work differently.

The Operating Model Problem Has a Long Pedigree

The consultant bringing the AI transformation deck in 2026 is doing something nearly identical to what the Agile consultant was doing in 2010. Both are offering a capability. Both are assuming the organization will reshape itself around that capability. Almost none of them do, because reshaping an operating model is a different and harder project than deploying a tool, and the two are routinely sold as the same thing.

The technology question and the organizational design question are separate questions. Buying AI licenses solves the technology question. It does not touch who decides what, how work flows from one stage to the next, what approvals are required at each gate, or how success gets measured. An organization that required six approvals to ship a feature before the AI rollout still requires six approvals after it. They are now generating AI-assisted briefs for each approval meeting, which is a different shade of the same color.

You can automate a process that does not work. You just automate it faster.

The decision structures, the information hierarchies, the accountability assignments — these are what determine whether a new capability produces value or produces activity. Agile made this clear for anyone paying attention. The organizations that got lasting value from it were the ones that changed how decisions got made, not the ones that ran the ceremonies most faithfully. The same logic holds now.

The Measurement Trap Runs on Schedule

One of the more durable failure modes in transformation work is measuring the wrong thing at the right moment. In Agile transformations, the most common proxy metric was velocity: story points completed per sprint. Organizations tracked it religiously. Velocity climbed. Business outcomes stayed flat. The research my co-author and I reviewed for Agile Sucks! (When You Do It Wrong) found that roughly 70% of Agile organizations reported positive metrics while producing no measurable improvement in business results. The metric looked like success. The outcome looked like before.

AI transformations are running the same measurement trap. The dashboard metrics are adoption rates: percentage of employees using the tool, number of prompts run per week, licenses activated versus licenses purchased. These tell you something about behavior and nothing about value. An organization can achieve full AI adoption in every sense the dashboard tracks and see zero improvement in anything that matters to the business. The harder question — what specific problem does this tool solve, in which specific workflows, and how will we measure the actual output — is being deferred or skipped in most deployments.

What Separates the Organizations That Actually Get Value

The research is consistent on what distinguishes organizations seeing real ROI. It is not vendor selection or training hours logged. It is whether the organization treated AI adoption as an operating model design problem and worked through it accordingly: start narrow, map the actual workflow, find where the real bottlenecks live, redesign the flow around the tool's specific capability, then measure what actually changes in the output.

This is not a novel methodology. It is what good transformation work has always looked like when it worked. The leaders who get this right ask the harder question before the announcement, when the operating model can still be shaped. They start with what the operating model needs to look like for AI to produce the result they want. Deployment at scale comes after that, if at all.

The answer to that question almost always involves changing something that someone with authority does not want changed. That is why most organizations skip the question. It is also why 95% of them are getting the result they are getting.

If you have run a transformation before under a different name, what did you learn from it that you are actually applying this time? That is the question worth sitting with before the next rollout plan goes to the board.