The caterpillar vs the butterfly. What true AI transformation looks like.

Deloitte just published its 2026 State of AI in the Enterprise report. They surveyed 3,235 business and IT leaders across 24 countries, director-level and above, all with direct involvement in their companies' AI initiatives. The headline number making the rounds on LinkedIn this week is that 66% of enterprises report productivity and efficiency gains from AI. That sounds like progress. But a different number in the same report tells a more honest story.

Only 34% say AI has actually changed how their business works.

That means two-thirds of enterprises are getting faster at the same processes they were running before AI showed up. They're doing the old work quicker. The other third is doing different work entirely, creating new products, reinventing core processes, rethinking business models. Both groups report productivity gains. Only one group is actually transforming.

The question the report raises but doesn't quite answer is: why? Same technology landscape, same vendor options, same model capabilities. What is the 34% doing that the 66% is not?

After spending fifteen years in enterprise pre-sales across fintech, Fortune 100, and education, watching organizations adopt (and fail to adopt) new technology at every scale, I think the answer has very little to do with technology and almost everything to do with what happens before the technology gets deployed.

The number nobody's talking about

84% of companies in the Deloitte survey have not redesigned jobs around AI capabilities.

Read that again. 84%.

They gave people AI tools and changed nothing about how the work itself is structured, who does what, or how teams operate. That is not an AI adoption strategy. That is an IT rollout with a bigger budget. And it explains the gap between the 66% and the 34% more clearly than any other data point in the report.

"Transition" and "transformation" both involve change, but they differ in scope, depth, and outcome.

A transition is moving from one state to another while keeping the core identity intact. A step in a journey, not a fundamental overhaul. A transformation is a deep change in form, nature, or identity. A caterpillar doesn't become a better caterpillar. It becomes something that operates in an entirely different way.

Transition is changing where you are. Transformation is changing what you fundamentally are.

The 66% transitioned. They moved existing workflows onto AI and called it adoption. Same meetings, same reporting chains, same decision-making process, just with a chatbot sitting next to it. The 34% transformed. They looked at the work itself and asked whether it should still exist in its current form before they put any technology near it.

What discovery looks like in practice

When I was working with PennyMac on eClosing adoption, the technology worked on day one. It was legally defensible, technically sound, and faster than paper. But adoption wasn't going to happen from a slide deck.

So I spent a morning sitting with their loan originators, watching their current process and listening to what frustrated them about it.

That did two things. First, the people who would actually use the tool felt like they were part of building the solution, not having it forced on them. They saw their own pain points reflected in the conversation, and when the solution showed up, it already had their fingerprints on it. That changes the emotional relationship between a team and a new tool in ways that no training program can replicate.

Second, when I walked into the stakeholder demo, I could speak their language. I could reference specific bottlenecks their originators dealt with at 2pm on a Tuesday. I could show exactly how the workflow would change based on what their teams were experiencing that day, not based on a generic use case from a slide deck. The demo landed because it was grounded in their reality, not in our product's feature list.

Discovery isn't a phase you skip to get to the demo faster. It's the reason the demo works.

And this is exactly what the 34% understood before the 66% did. You cannot transform an organization's relationship with technology if you haven't first understood the organization's relationship with its own work.

The discovery gap

Deloitte's report identifies insufficient worker skills as the biggest barrier to integrating AI into existing workflows. I think the framing is off.

What most organizations actually have is a discovery gap.

They bought AI tools without running discovery on their own processes first. They skipped the part where you sit with the people doing the work and ask: what does your Tuesday actually look like? Where does information get stuck? What would you change if you could, and what are you afraid of losing?

Those aren't skills questions. Those are the questions a good solutions engineer asks in the first 30 minutes of a discovery call, and they're the same ones that should happen before any AI deployment reaches a single end user. The fact that 84% of enterprises haven't redesigned jobs around AI tells you those questions never got asked, or they got asked by the wrong people at the wrong time.

The report's own data supports this. 53% of companies say their primary talent strategy adjustment is educating employees to raise AI fluency. Only 33% are redesigning career paths. Only 30% are combining or reimagining organizations based on new work patterns from AI. The majority response to a transformation challenge is a training program. That's a transition move applied to a transformation problem.

Governance as discovery

The Deloitte report also found that enterprises where senior leadership actively shaped AI governance achieved significantly greater business value than those delegating the work to technical teams alone. Only 21% of companies report having a mature governance model for autonomous AI agents, despite 74% planning to deploy agentic AI within two years.

That tracks with everything I've seen. Governance isn't a compliance checkbox you fill out after the deployment is live. It's the organizational equivalent of running discovery before you build the demo. You're deciding where humans stay in control, what gets automated, who's accountable when something breaks, and what the escalation path looks like when an AI agent does something unexpected at scale.

The companies getting governance right are the same ones getting adoption right, because both require the same discipline: asking hard questions about how work actually happens before you change how it's done. The 34% treat governance as a design conversation. The 66% treat it as a risk and compliance exercise they'll get to eventually.

What this means on Monday morning

If you're reading this and recognizing your organization in the 66%, the move isn't to buy different AI tools or hire more data scientists. The move is to run discovery on your own organization before your next deployment.

That means sitting with the people who will use the tool, not the people who bought it. It means mapping current-state workflows before designing future-state ones. It means asking what people are afraid of losing, not just what they're excited about gaining. And it means treating governance as a strategic capability that lives alongside your AI roadmap, not underneath it.

The 34% didn't have better technology. They asked better questions before deploying it.

---

For those who made it this far, three questions I wrote down while thinking about this:

1. If you paused your AI rollout tomorrow and spent two weeks just interviewing the people who are supposed to use it, what would you learn that your implementation plan doesn't account for?
2. When your team talks about the "skills gap," are they describing a training problem or avoiding a harder conversation about how work needs to change?
3. Who in your organization has permission to say "we deployed this wrong" without it becoming a career risk, and what happens if the answer is nobody?