Let me start with a confession.
I use AI every day. So do you. We draft with it. We summarise with it. We tidy our inboxes, polish our reports, prep our meetings with it.
And here is the uncomfortable part.
Most of us are doing exactly the same work we did two years ago — just a little faster.
We picked up a remarkable tool. And we pointed it at our old habits.
Adoption vs. reinvention
I want to draw a line between two words that look similar but lead to completely different places.
Adoption is when you bolt a new tool onto an old process. The email gets written faster. The slide gets made quicker. But the process underneath? Unchanged. The shape of the work is exactly what it was.
Reinvention is when you stop and ask a harder question — not "how do I do this task faster?" but "should this task exist at all?"
Adoption asks what the tool can do. Reinvention asks what the work should be.
Try this yourself. Think about the last time you used AI at work. Did you just do the same thing faster? Or did you actually change what you did, or whether you needed to do it at all?
For most of us — and I include myself here — it's the first one. And that gap, between using AI and rethinking the work, is what I call the reinvention gap.
What pilot purgatory actually costs
This isn't abstract. There are numbers.
PwC's 2026 AI Performance Study found that around 74% of all the economic value AI is creating is being captured by about 20% of organisations. Three-quarters of the value, flowing to one in five.
What do those 20% know? Not a better model — they're running the same ones the rest of us are. The difference: they were about twice as likely to redesign their workflows around AI rather than bolt it onto the process they already had.
The leaders didn't buy a better tool. They redesigned the work.
Everyone else is stuck in what I call pilot purgatory. McKinsey looked at organisations rolling out AI agents — the autonomous kind that actually do the work. Two-thirds of enterprises have tried them. Fewer than one in ten have scaled them to anything that delivers real value.
Two-thirds in. Under ten percent out the other side.
And here's the thing that should reframe how you think about this. When McKinsey asked what was actually blocking them, around eight in ten pointed at the same thing. Not the model. Their data. The information was scattered, locked in systems that don't talk to each other, inconsistent, or simply not trusted enough to let an agent act on it.
Pilot purgatory isn't a technology problem. It's an operating-model problem wearing a technology costume.
Three throughlines
So what do you actually do about it? Three ideas I keep coming back to.
One: redesign the work, not the tools. Don't ask what AI can do for your task. Ask what AI means for your work. Those sound similar — they are not. One makes you a little faster. The other changes the shape of the job. BCG now treats work reinvention as a CEO-level mandate, and their advice is blunt: remove the unnecessary work before you digitise it. Otherwise you're just paving the cow-path.
Two: your advantage was never the model. McKinsey estimates roughly nine in ten organisations now use AI in at least one function. When everyone runs the same model, the model is no longer your advantage — it's the price of entry. The real advantage lives in three things: your proprietary data, your redesigned workflows, and your distinctive people. Data. Workflows. People. Notice what's not on that list.
Three: agents demand new accountability. The new generation of AI doesn't just do tasks — it runs for days, holds a goal, makes decisions. And here's the uncomfortable finding from Harvard Business Review: when AI was framed as an "employee" rather than a tool, managers caught 18% fewer errors and personal accountability fell by nine points. The moment we called it a colleague, we trusted it more and checked it less. An agent can't be accountable. It can't be hauled in front of a committee. Only a person can.
The university altitude
The same gap runs through institutions, not just individuals.
Students are using AI to do the degree. 94% now use generative AI for assessed work, according to this year's HEPI survey. But the degree itself — what we teach, how we assess it, why anyone should pay for it — has barely moved. That is adoption at the level of an institution. And it's the more dangerous version of the gap.
ANU research finds public confidence in Australian universities has fallen from around 79% in 2019 to about 62% now. Employers are loudly going "skills, not degrees" — around 85% say they hire on skills. Meanwhile, the knowledge we used to hold exclusively is now available for free, at 2am, in any student's pocket.
A degree's relevance no longer comes from the knowledge we transmit. It comes from everything AI can't give: judgment under uncertainty, formation as a professional, mentorship from someone who has done the thing, a credential the world trusts, and a human community learning together.
Five questions for Monday
The reinvention won't be written by people who talk about it. It will be built by the people who run the place — the professional staff who design and deliver the systems, processes, and services that make an institution actually work.
You don't need a transformation programme to start. You need five questions. Take them to your team:
- What in my work exists only because of an old constraint — and could now just go?
- If I redesigned this around AI instead of bolting AI on, what would it actually look like?
- What here is genuinely ours — our data, our judgment, our relationships — that no one else can copy?
- Where an agent now does the doing, who is accountable for the deciding? Name the human.
- What's the one end-to-end journey where getting this right would matter most?
You won't answer all five by Monday. That's fine.
Asking them is the shift.