How do you use AI in a real business?

As a programme, not a pilot. I lead a multi-year programme at Rapha to roll AI out across the business, built on Google Cloud. What is in production today is not one chatbot: it is multiple agents and workflows, some internal, some customer-facing, and some fully autonomous, running end to end without a person in the loop. The constraint that shaped all of it was the brand. Rapha is a premium cycling apparel brand, so anything that sounds like a generic assistant is a cost, not a saving.

Internally, it is tooling and workflows that take real work off people: the repetitive, well-defined processes where an agent can be trusted to act, and the heavier ones where it prepares the work and a person decides. Customer-facing, it is agents that help customers directly, held to the same standard of experience as anything else we put in front of them. I will not list the specific use cases here, but both sides are live, not planned.

It has changed how we build as well as what we build. I have rolled AI into developer experience: agent skills, tooling and new ways of working for the engineering team, so the same leverage we give the business applies to how we ship. I build agents myself, on Vertex AI Agent Engine and with MCP servers, and I still deploy production code, so I know what the tools can and cannot do before anyone promises it. The same discipline applies as to the rest of the estate: integration patterns, failure modes, observability and security, with PCI DSS and GDPR in the room from the start.

Clearly, because it is a real one. I do applied AI and agent work: putting models to use inside processes and products, grounded in the data platform I built on GCP and BigQuery. I do not do data science or ML modelling. If you need someone to train models, that is a different hire. If you need someone to get AI working safely across a real business, with the commercial case made in cost, forecast and return, that is the work I do now.