Building an Agentic Company
What actually happens when AI runs your operations. Delivered at AND Digital FutureSkills, Pentahotel Reading, on 17 June 2026.
About the talk
Tom Dean built an AI agent — L1 — to run the operations of his consultancy: inbox monitoring, CRM management, daily briefings, meeting capture, and prospect tracking. This talk documented what happened over the first three months in production. Real failures, permanent fixes, and genuine business value.
The three failure types
- Context failure: the agent replied to the wrong thread, splitting a booking into three separate email conversations
- Verification failure: calendar times were one hour wrong during British Summer Time — meetings nearly missed
- Persistence failure: tasks listed in the evening vanished overnight when the session reset at 4am
Key takeaway
The moat is not the model. Everyone has access to the same models. The moat is the operating layer you build around it — skills-based design, persistent memory, verification steps, and feedback loops that turn failures into permanent improvements.
About the event
FutureSkills is AND Digital's programme of events for technology and business professionals. AND Digital is a digital transformation consultancy with offices across the UK. The Reading session was held at Pentahotel Reading.