We were in a review meeting when my boss said it. I'd just walked the team through a piece of work that had come back faster and cleaner than anyone expected, and he sat with it for a second, and said: "I didn't realise AI could do the doing."

He'd heard all the noise. There is an enormous amount of it about how incredible AI is, and almost all of it is about the talking. Drafting the email, tidying the report, summarising the meeting. The edges of the job.

A friend told me this week she's switched to Claude. She's a smart, capable person, and she's using it, in her words, exactly the way she used ChatGPT: to write her emails. That's where most people are, even the ones who have moved to the better tools. The problem isn't that they don't believe AI is powerful. It's that nobody has shown them how you get from "write my email" to AI doing the actual work.

Here's what the middle of the job can look like now.

A charity needed to move decades of history out of one CRM and into another. More than 400,000 gift records, years of donations, pledges and relationships, all of it tangled up in the quirks of a system nobody fully remembered building. The kind of project that has a six-month figure written next to it before anyone even starts.

We did it in about six weeks.

Not by cutting corners, and not by handing it to a machine and hoping. I built AI into the work itself. The cleansing that would have taken weeks of eyes on a spreadsheet, done in passes I could check. The SQL I'd normally hand-write line by line, generated, tested and refined in a fraction of the time. The field mapping, the transformation, the quality checks, the documentation, all of it moving at a speed that used to be impossible for one person.

The expertise didn't go anywhere. I still had to know what "right" looked like. Which records to trust, which to question, where the data was lying, what the client actually needed on the other side. AI doesn't make those calls. It does the typing, so the judgement can go on the judgement.

My friend on Claude and that charity with its 400,000 records could be using the very same model. The difference is knowing what's possible, and knowing how to build it into the work.

For anyone running an organisation, this is the part worth knowing. The migration you've been putting off. The cleanup you've been meaning to do for three years. Those things sit undone because they've always been too big for the people you have. That maths has changed. One experienced person, with AI built into how they work, can now deliver what used to need a team and a timeline you couldn't justify.

If you've got one of those jobs sitting on a shelf because it's always been too big to start, it might be worth another look. I'm happy to talk it through.