Now

Can we build AI that pays attention rather than demands it?

The attention economy taught a generation of products to interrupt, to escalate, to mistake engagement for value. What I'm pushing toward - and I think it's the more interesting frontier - is the attentive economy: systems that notice, that hold context lightly, that know when not to speak.

The shift is small in language and enormous in consequence. It asks the model to be curious rather than confident, to admit what it doesn't know, to leave room for the person on the other side of the screen. I've been calling this nescient AI - principled not-knowing as a design stance - and it keeps opening doors I didn't expect.

The work now is trying to do this at scale. Inside government, with GDS, that means agentic services that meet citizens where they are rather than where the org chart says they should be. Outside it, I am building, and tending to, a number of agent projects - a slower, more attentive sensibility that I hope to say more about when I'm ready to share.

Domestically, there's a lot happening. My boys are grown and growing. It is A-levels season for one and back home from Uni for the other. My home is still half-renovated and probably will be for a while. I'm reading less and making more than I expected to this spring, and finding that the making is doing some of the thinking I used to ask the reading to do.