Almost nothing on stage was about the cloud, and almost everything was about agentic AI. That is not a complaint so much as an observation about where the centre of gravity has moved. The infrastructure conversation that dominated these events for a decade has become the assumed substrate; the interesting questions are all a layer up, about what you build on top of it and how you let software act on its own.
Listening to the case studies, a clear divide emerged between two kinds of organisation. The first asks what they would build from scratch if the legacy did not exist, ring-fences the old estate, and reimagines on top of it. The second does the opposite — it wraps expensive agents around workflows that should never have existed in the first place, and is then surprised when the results are underwhelming.
That distinction matters more than any particular model or vendor. An agent dropped onto a broken process inherits the brokenness and adds cost. The same agent, pointed at a process you have had the discipline to redesign, compounds. The technology is rarely the constraint. The willingness to question the shape of the work is.
The other number that stayed with me: only around 8% of global AI spending happens in Europe. I do not think that is principally a money problem — it is a mindset one. I heard one bank framing AI as a way to cut costs and another framing the identical technology as a way to grow revenue. Same tools, opposite ambitions, and you can already guess which institution will look more interesting in five years.
So if I had one suggestion for the organisers, it would be to rename the event — and for the rest of us, to stop treating “adopting AI” as a procurement exercise. The leverage is not in buying the agents. It is in being brave enough to redesign what they operate on.