I was at a leadership offsite last month — one of those half-day things where you eat too many muffins and the conversation eventually gets real. Good group of owners. Someone mentioned they'd just finished rolling out AI tools across their team. Big investment. Six months of work. The room murmured.
"How's it going?" I asked.
"Amazing," he said. "We're doing the same work about 30% faster."
He looked pleased. I tried not to wince.
Because here's the thing. Thirty percent faster at yesterday's work is not the game. It feels like progress — and in a narrow sense it is — but it is not the move that's going to matter in two years. You've bought yourself a faster horse at the exact moment your competitor is figuring out how to buy a truck.
I've started calling this the photocopier trap. Remember when colour photocopiers arrived? Great machines. Changed nothing fundamental about how a business actually worked. Everyone bought one, costs came down, and two years later your competitive position was roughly where it started — except you'd spent the money.
A lot of NZ businesses are sitting in exactly that spot with AI right now. Using it. Happy it exists. Not changing anything underneath.
The Question Most of Us Aren't Asking
The conversation worth having — and I'll be honest, I've had to drag myself to have it with my own businesses too — is not "how do we make our team 20–30% faster?"
The question is: what becomes possible that wasn't possible before, when intelligence is cheap and can run autonomously?
That's a different question. And it lands in a different place.
We're not just talking about drafting emails faster or summarising reports. We're talking about agents that can own an end-to-end process — researching, drafting, cross-checking, flagging exceptions, routing to humans only when it actually needs a human. Not a person touching every step. That's not efficiency. That's a different org.
The businesses getting ahead right now are treating this like an R&D budget, not a stationery order. Real money ring-fenced. Permission to pull apart one process and rebuild it from scratch. Tolerance for the experiment that doesn't work.
For a $2–$20M business, that looks like $100k--$500k a year set aside for genuine experimentation. Not buried in the IT line. Not squeezed out of an existing project budget. An actual allocation, with someone responsible for it.
Datacom's 2025 State of AI Index tracked this across NZ organisations. Eighty-seven percent already use AI in some form. Eighty-eight percent of those say it's benefited their business. But only 12% have scaled it across their whole operation. That gap isn't a technology problem. It's a redesign problem. And redesign requires a different kind of commitment than buying a licence.
Three Ways I've Seen This Play Out
These aren't case studies. They're composites — patterns I've watched repeat across firms in the last 12 months. If you recognise your business in here, that's the point.
Picture a Wellington accounting practice. Solid firm, respected partners, around $4M in fees. They give everyone ChatGPT Enterprise in early 2025. Tax returns done 40% faster. Compliance notes polished overnight. Easy wins, and fair enough.
Six months later, the partners have two new problems. Quality is inconsistent because nobody touched the review process. And clients are asking for an AI discount. First-order gains, none of the second-order ones. Staff still doing the same handoffs. Org chart still identical. One partner, when I sketched this out for him: "Yeah. We treated it like a new photocopier."
Now picture an Auckland strategy consultancy. Twenty people, around $9M. They set aside $250k as a genuine AI lab budget in mid-2025. Not tools — experiments. They pick one service line, due-diligence packages for private equity deals, and give a small squad actual permission to tear it up. A consultant, a junior, an outside AI engineer, and a mandate to fail. Ninety days later they've built an agentic workflow that pulls public data, drafts 70% of the report, flags risks, and routes exceptions to humans. Cycle time: six weeks to ten days. Same headcount, twice the mandates. The owner: "The budget wasn't for tools. It was for failure tolerance. That's what paid for itself in the first quarter."
And the third pattern — the one that keeps me up a bit. A Christchurch engineering practice, around $12M, watches their larger competitors roll out AI-assisted design review. Sensible people. They decide to wait and see what actually works before committing. By the time they move in late 2025, the leaders have already redesigned their teams around agentic review loops. The follower saved the R&D spend. And gave it back in lost tenders. "Exponential improvement means the follower gap closes in months, not years." That one sticks.
Two Jobs. One's Getting Repriced.
Here's the thing I keep coming back to for businesses like ours. Professional services — accounting, law, consulting, engineering, advisory — we've always been doing two completely different kinds of work. We bundle them together and send one bill. That bundling is quietly coming apart.
The first is execution. Drafting, researching, structuring, cross-referencing, data extraction, compliance review. The mechanics of producing good work. These skills took years to build — genuine craft. I'm not dismissing that. But the market is repricing them regardless of what we think about it. AI produces a legally competent first draft, a technically sound compliance review, a well-structured report — faster and cheaper than any of us. That doesn't diminish what it took to learn those skills. It does change what clients will pay for them.
The second is harder to name. Call it judgment.
Judgment isn't just reviewing the document. It's understanding what the client is actually trying to achieve — as opposed to what they've asked for. It's knowing how a particular counterparty behaves across three prior deals — what they'll concede, what they won't give on principle, and which clause they're pushing not for legal reasons but as leverage somewhere else. It's reading that your client is under board pressure and this transaction needs to close fast for reasons they haven't said aloud.
It's asking the question they didn't know to ask. Reframing the problem before solving it. Seeing the whole picture — the business, the relationships, the trajectory, the risk — and giving advice that fits the situation they're actually in, not just the document in front of you.
It's what allows me to write these articles.
Execution, AI will handle. Full stop.
Judgment? Still yours. And here's the upside: when you're not buried in execution, your judgment goes a lot further. You can serve more clients. You can take on more complexity. You can charge for what you're actually worth — which is the insight, not the hours.
The structural advantage available to professional services firms right now isn't efficiency. It's a genuine repricing of where the value sits. But only if you're willing to unbundle.
The Question for Your Next Offsite
Here's the one I'd put on the whiteboard: If our three nearest competitors all moved to an agentic operating model in 2026, how exposed is our current setup?
If that makes you a bit uncomfortable — good. That's where honest thinking starts.
And the practical one: what is one end-to-end process you could actually redesign around agents in the next 90 days? Not investigate. Not agenda-item-it. Redesign.
We still have the window. Not wide open, and not for much longer — but we have it. The owners who treat 2026 as the year they build a genuine AI-native R&D muscle, rather than buy more licences and call it transformation, are the ones who'll be writing the next chapter of their firms.
The rest will still be faster at yesterday's work.
Good luck out there. And don't forget to have fun!
