The price of a single AI response keeps falling, yet the invoices keep climbing. Agents think in long chains, call other agents and run around the clock, and many organisations are discovering their annual AI budget was gone well before the year was over.

This opening keynote explains where the money really goes, who should own the bill, how leaders set limits without slowing their teams down, and how to explain to a CFO why success can look so expensive.

Here is a simple test for any AI programme. If your main model provider doubled its price, changed its terms or could no longer keep your data where your customers expect it to stay, how long would it take to move? For many organisations the honest answer is months, because prompts, evaluations and workflows were all built around one vendor without anyone deciding that on purpose.

For British organisations the question is sharper still, as customers and regulators ask where data is processed and which jurisdiction a provider answers to, and as the UK invests in its own sovereign AI capability. Our panellists share how they run switching drills, which parts of the stack they insist on owning, and the contract clauses that protect their exit.

The phone call never died. It is still the channel customers reach for when something truly matters, and a new generation of voice AI can now follow context, tone and intent across an entire conversation.

In fifteen minutes, this session explores where voice AI already works, how it copes with regional accents and customers who switch languages mid-sentence, and how to keep trust when the voice on the line is not a person.

Europe pushed back the AI Act's high-risk obligations to December 2027 for many standalone systems, while the UK has chosen sector regulators and sandboxes over a single AI law. For organisations working on both sides of the Channel, that means two very different playbooks, and the EU's transparency duties are already live.

Our panellists share how they are using the extra time wisely, which systems they are prioritising, and how responsible AI practice is becoming a reason customers choose them rather than a burden they carry.

Most AI assistants still start every conversation as strangers. They forget the customer they spoke to yesterday and the decision the team made last week. A new generation of agents can carry memory across tasks, learn how a team works and pick up a project exactly where it was left.

This talk explores what memory makes possible and what it demands: which memories an enterprise should keep, share or deliberately forget, and how to protect privacy under UK GDPR when your AI never truly closes the file.

Your agents are starting to call tools, negotiate with other systems and hand tasks to agents built by partners and vendors. Behind every one of those conversations sits a new layer of protocols and connectors that most enterprise architectures never anticipated.

This session demystifies the emerging standards that let agents work together, and shows how to adopt them with clear rules for trust, logging and accountability from the very first day.

Most enterprises now have a long list of agents and a much shorter list of agents that earn their keep. Pilots linger because nobody wants to be the one to end them, while the few that work wait for the data, controls and funding they need to grow.

Our panellists bring real, anonymised projects to the room, and the audience votes on each one before hearing what actually happened. An honest, lively look at what to measure, when to stop and how to grow what works. The results are revealed at the close.

The race for ever larger models makes the headlines, but inside many banks, insurers and asset managers in the City of London a quieter story is unfolding. Smaller models trained for a single domain are proving faster, cheaper and easier to govern.

In fifteen minutes, this talk shows where the specialist beats the giant, what it does to accuracy and cost, and how to decide which problems truly need the biggest brain in the room.

Every organisation has a few people who simply know. The engineer who can hear when a machine is about to fail, the underwriter who senses a bad risk before the numbers show it. Many of them learned by doing junior work that AI now does, and many are approaching retirement.

This closing conversation explores how leaders use AI to capture that expertise while those experts are still in the building, and how they make sure the next generation still learns the craft.