An interactive AI avatar works when it has a role, approved knowledge, a goal and a clear next action for the user. Ben English of Kaltura taught that method in a hands-on workshop at aiblLIVE London 2026. Tables designed their own agent on paper before configuring one.
This free guide turns the session into a method you can run with your own team, whichever agent builder you use.
Make four decisions before you build anything: the agent’s role, what it knows, the goal it works towards and the action it offers at the end. In Ben’s demo, the goal was a meeting booked with a human. Other goals include a passed test, a share of training videos watched or a captured email address.
This AI has a role, it has a purpose, and it has knowledge.
Ben English, Director of Solution Engineering, AI Specialist, EMEA and APAC, Kaltura
Give it only approved content and let it answer from nothing else. Ben described this as retrieval-augmented generation. The agent finds the relevant passage in your content, then the language model turns it into a reply. It does not go to the open web. Content that changes often needs a named owner.
Tell the agent what the next step is and let it offer it inside the chat. In the demo, it answered a question, generated next-step buttons, pulled a short video clip and registered the visitor for an event. Once it judged them a serious buyer, it booked a meeting with an expert into their calendar.
Write them into the agent’s playbook, along with its audience and opening phrases. You can ask for a calmer or more energetic tone. Off-limits topics depend on the setting: a customer website should avoid politics and religion, but a university that teaches both needs them left on.
What should the agent do? What does it need to know? And what should happen next?
Ben English, Director of Solution Engineering, AI Specialist, EMEA and APAC, Kaltura
Ben English is Director of Solution Engineering and an AI specialist for EMEA and APAC at Kaltura, the video and streaming company now building conversational AI avatars. He ran this workshop on the Customer track at aiblLIVE London on 20 October 2026.
If you want an agent that does more than answer questions, this is a method you can run next week. The full PDF goes step by step.
For customer experience, marketing, events, learning and support leads. Seven pages, about a nine-minute read.
