McKinsey reportedly found 10% of users consume 65% of its AI tokens.

That makes cost control important. But “use fewer tokens” feels like the wrong objective.

£100 of AI removing £1,000 of work may be great value.

www.businessinsider.com/mckinsey-par...

I’m increasingly wary of “What’s our AI strategy?”

Not because AI isn’t strategic, but because it makes the technology the objective.

A better question: what can we now make possible that we couldn’t before?

AI matters because it moves the boundaries of what an organisation can do.

There’s an odd contradiction in recruitment. We increasingly want people who are good at using AI at work, then assess what they can do without it.

Maybe we should give candidates the tools and assess how they use them.

AI might ultimately make recruitment more human, not less.

I like the principle behind Debian’s new AI policy: use the tools, but you’re still responsible for what you produce.

As AI becomes normal work, perhaps governance needs to focus less on “did you use AI?” and more on “can you stand behind the result?”

www.theverge.com/tech/986789/...

There's a long-running critique of how appraisal works in the public sector, and I think it connects to something I keep bumping into on the technology side: business cases tend to favour benefits that arrive quickly, can be seen, and can be counted in cash.

I started coding partly because it was good for my head.

A bounded problem. Contained logic. A right answer somewhere at the end of it, and a couple of hours where nothing else got in. The friction was part of the point — being stuck, working it out, the small release when it finally ran.

A question I keep turning over: if you work inside a command-and-control culture, can you still run your own function in an empowered, open way?

I like continuous improvement. But it still needs an investment test.

At some point, the next improvement costs roughly the same but delivers less value than the last.

Maybe mature product management isn't just knowing what to build next. It's knowing when enough is enough.

I’m a self-taught coder. I’ve built about 10 apps now and increasingly I don’t even look at the code.

AI hasn’t made me an AI developer. It’s changed what I’m capable of building — and how quickly I can test an idea.

That distinction feels more important than a lot of the “AI use case” debate.