Recently, I finished working through the last of three sets of survey responses I had collected over two years. They came from separate client engagements and included 137 people: 29 CEOs, seven members of an internal AI task force, and 101 employees inside a single organization.

Ultimately, I didn't design the surveys so they could be compared. The questions were different, and so were the purposes—not to mention that it was client work and not a formal research study.

Nevertheless, I thought it would be interesting to look at the responses next to one another and see what emerged.

What Changed and What Didn't

What became immediately clear was that the technology had changed considerably since I fielded the first survey, but the organizational work of transformation had not.

In late 2024, most conversations about generative AI were framed around a chatbot that did, more or less, one thing. Since then, models have become multimodal, learned to use tools, and started carrying complex work through multiple steps over a long period of time. In short, we've officially entered the agentic AI era.

However, the questions inside organizations have remained almost exactly the same: Where should we use AI? What information can we trust it with? How do we train people? How do we integrate it into existing work? How do we keep experimentation from becoming another collection of disconnected tools and individual workarounds?

In other words, the technology has advanced, but not much else has.

The View from the Top

Unsurprisingly, of the 29 CEOs I surveyed in December 2024, 28 selected operational efficiency as a primary goal for AI. Of note, though, was the distance between where those leaders wanted to go and where their organizations stood. More than half were still in the earliest stages of adoption. Another third were experimenting. And more than two-thirds had no formal governance in place.

The leaders named barriers like privacy and security, limited expertise, and the difficulty of integrating new tools into existing processes. It was less a lack of ambition than it was a lack of architecture.

One CEO described a “lack of a strong process-driven mindset” that left AI use “ad hoc and inconsistent.” Another wanted “a consolidated corporate approach versus individuals going at it.”

What they were getting at, without knowing it, was that it's ultimately structure and governance that create the conditions for AI transformation to happen. The technology is important, but it's second to having the environment for the technology to be able to do its thing.

Experimentation Is Not Integration

Around that same time, I surveyed seven members of an internal AI task force at another organization. The difference is that these were not executives considering the potential of AI. They were the people in the actual day-to-day work and they were already using it.

In that survey set, just like the one before it, not one single person described AI as well integrated across departments. Every respondent said it was either poorly integrated or only somewhat integrated.

One person described AI as the “bright shiny object” and worried expectations had outrun what the tools could deliver. Others said learning to use AI could add to someone’s workload before reducing it. The insight was that the group still needed practical skills, clearer policies, time to learn, internal champions, and a way to evaluate what was working.

What we know now is that this is the middle of transformation, and it’s usually where the work becomes most difficult. Access to tools creates activity, but it’s doesn’t create integration. Integration requires decisions about where experimentation should happen, how results will be judged, and how a useful discovery becomes part of the way the organization operates.

Just as the executives in the first group had sensed, the missing layer here was governance: the structure that turns individual experimentation into an integrated way of working.

The Work Has to Be Ready

In my most recent survey, I asked 101 employees about how work actually moved through their organization.

More than half struggled to find information across documents, email, systems, and files. A similar number wanted faster answers from internal policies, manuals, and project files. Others pointed to changing project details, missing information, repeated data entry, and work that moved through the organization without a clear source of truth.

The follow-up interviews I conducted made those frustrations more tangible. One leader needed at least three systems to invoice a small job. Another described various people completing the same work in different ways. Across the conversations, people talked about incomplete job information, inconsistent repositories, weak handoffs, and uncertainty about which information they could trust.

The interesting thing is that none of these are necessarily AI problems. They’re data problems and process problems and problems that many businesses face when they begin to scale.

What’s more, it's important to remember that AI can only retrieve knowledge the organization has captured. It can only integrate with processes that are clear enough to follow. It can only improve work the organization can see, understand, and evaluate. If the information is scattered and the process changes from one person to the next, adding AI may simply make the confusion move faster.

A Consistent Story

Seen together, the three surveys tell a consistent story. The CEOs had ambition but lacked the structure to pursue it. The task force had access to the technology but lacked a mechanism for turning experimentation into shared practice. The employees could see exactly where the underlying work was fragmented, but they had neither the tools nor the governance to transform.

What each group needed was an organization capable of absorbing what the technology could do—establishing governance and decision rights, making organizational knowledge accessible and trustworthy, creating enough consistency in the work for AI to become part of it, and building a mechanism for capturing what people learn and feeding it back into the organization.

These are not preliminary steps that happen before AI transformation. They are the work of AI transformation.

The models will keep changing, and the distance between what AI can do and what most organizations are prepared to do with it will continue to grow. Leaders cannot control the pace of technological development, but they can control whether their organizations are structured to learn, improve, and adapt as the technology advances.

The technology has moved. It's time for organizations to move as well.