In 2025, Project NANDA’s preliminary report on enterprise generative AI drew on more than 300 public initiatives, 52 organizational interviews, and 153 senior-leader survey responses. It reported that 95% of organizations were seeing no measurable return.
The number is what made headlines. What I keep coming back to is the explanation underneath it: the report highlights brittle workflows, limited contextual learning, and poor fit with day-to-day operations. My takeaway is about leadership, structure, and who has the authority to change how the work gets done.
I've sat inside efforts like this as the executive accountable for making them work, more than once, in more than one company. The pattern held with uncomfortable consistency long before I saw MIT's version of it in print. I stopped believing this was about which vendor got picked a while ago.
Starting from the wrong end
Almost everyone talking about AI right now is answering some version of one question: what can it do? Fewer people are asking the one that actually determines whether an organization benefits from it: what does this require us to become? Most companies get the sequence backward. They start with the technology's capability and work outward — buy the tool, run the pilot, hope the organization catches up around it. Strategy, leadership alignment, and the actual design of the work should come first. Technology should follow that, not lead it.
The quieter mistake sits right next to the first one: treating Future of Work as a workplace question — hybrid schedules, an HR initiative, a rollout plan handed to IT. It isn't. It's an enterprise-design question: how people, technology, organizational systems, and leadership have to evolve together to create meaningfully more value. Reduce it to policy, and you'll optimize something that was never the actual problem.
The distinction most AI strategy skips
Here's the piece I think gets missed most often: capability and authority are not the same thing.
Capability asks can it decide? Authority asks should it decide? A model can already draft the memo, price the deal, or flag the risk. None of that tells you whether it should be the one framing what the memo needs to argue, what the deal is actually for, or which risk is worth taking. Machines can become extraordinarily capable without ever holding legitimate authority — and the more capable they get, the more that distinction matters, not less. I call this Intelligence Stewardship™: as high-quality answers become abundant, human value has to move from producing them toward framing the right problems, asking the questions worth asking, evaluating what comes back, and being willing to own the outcome. None of that is a new skill. It's the oldest work of leadership. What's new is that there's no longer enough production work left to hide behind instead of doing it.
Our task was never to stay superior to AI at everything. It's to become better stewards of what intelligence now makes possible.
The fair objection
Here's the honest pushback, and it deserves a real answer: if the barriers are organizational, doesn't that let the technology off easy? Maybe some of these tools just aren't as good as the demo, and "the organization wasn't ready" is a convenient way to avoid saying so.
The same report offers a useful clue. In its sample, external partnerships using learning-capable, customized tools reached deployment roughly 67% of the time, compared with roughly 33% for internal builds. These were self-reported outcomes, and the authors caution that the difference does not establish causation. To me, the finding reinforces the importance of integration, ownership, and authority to adapt workflows. It does not mean model quality is irrelevant.
What I actually build
I don't think the goal is ever one successful AI rollout. Organizations that simply add AI to an operating model they never redesigned fall behind the ones that build the capacity to keep redesigning themselves as what's possible keeps changing. I call that capacity Repeatable Transformation™, and I think it has to become a core organizational competency in its own right — not a one-time project with a vendor and an end date.
That's the work I built Whole Enterprise™ to do. I work as an embedded advisory partner with executive teams — an ongoing rhythm, not a single engagement that wraps and closes — across the six challenges every organization eventually has to move through to become genuinely future-ready: leadership, strategy, AI and automation, workforce and skills, organizational design, and culture. Each builds on the one before it. Skip leadership and strategy to get straight to the AI pilot, and you risk repeating the pattern I’ve seen in practice.
The question worth asking this week
Before the next AI initiative gets approved, one question tends to cut through the noise faster than a strategy deck:
Where in your organization has capability quietly started acting like authority — and did anyone actually decide that, or did it just happen?
If you can answer that crisply for every function AI now touches, you're ahead of most of the executive teams I sit with. If you can't, that's not a failure. That's just where the real work starts.
This is the founding conviction behind Whole Enterprise™: AI should expand what people are capable of without quietly taking over who gets to decide. If your organization is further along on adoption than it is on judgment, I'd like to hear how you're thinking about closing that gap.