The AI Summit London 2026

On the Applied AI stage at Tobacco Dock, I joined a panel on what it actually takes to get AI tools into the workflows that run a business—CRM, finance ops, customer support, internal tooling—rather than leaving them stranded in pilots and demos.

Moderated by Omdia's Adam Holtby, the session paired practitioner views from Centrica's Wendy Clack and Celigo's Phil Doran with my own perspective from the infrastructure side. The through-line: the model is rarely the hard part. The hard part is integration, data plumbing, and the change management needed to get people to trust a tool that quietly rewires how they work.

My argument was that the best deployments make AI invisible. Users shouldn't have to think about which model is running or where inference happens—the value shows up as a faster quote, a cleaner reconciliation, a resolved ticket. That only holds when the tool sits inside existing systems, respects the latency and governance realities of the enterprise, and fails gracefully when it isn't confident.

We also dug into why pilots stall: unclear ownership, brittle integrations, and a habit of measuring model accuracy instead of workflow outcomes. The consensus was pragmatic—start with one narrow, high-friction process, instrument it properly, and earn the right to expand.