Plenty of AI, as you can imagine. We got to shake hands with product leads from Google and Anthropic, alongside some genuinely brilliant talks that had nothing to do with AI at all. A big thank you to Stuart Clarke MBE, Deb Hetherington and the Leeds Digital team, everyone who opened their doors, and Venture Community for giving us the stage.

And it isn't over yet. Leeds Digital Festival runs until Thursday 2 October, so if you've spent the week experiencing festival FOMO from afar, there's still time to fix that.

That matters because the economics of putting capable AI to work are changing very quickly. The intelligence itself is getting cheaper, which means more of the challenge shifts to everything around it - the information it needs, the systems it connects to and the work you actually want it to do.

The cost of intelligence is falling fast. The valuable bit is increasingly what you build around it.

Australian Defence Minister Richard Marles put it rather neatly: the agent asked a question, was refused, and then “scaled the fence”. OpenAI says the model took actions it did not intend. The interesting bit isn't really that an AI “hacked” the government. It's that the agent had a goal, hit an obstacle and decided the obstacle was something to solve.

Anyone putting an agent to work needs to define where it should stop as carefully as what it should achieve. The more capable they become, the more important those boundaries get.