The origin story
Our story
How two blokes in a Wetherspoons pub ended up with an AI co-host.
How Rob and Richard met
Rob and Richard have known each other for years, originally crossing paths through the school run. Rob founded Cogendo, the company behind PerformanceHub, and is happiest somewhere between writing code, experimenting with AI and finding unnecessarily clever ways to avoid doing anything manually. Richard spent years as a senior director in technology consultancies, giving him a finely tuned instinct for which ideas might actually survive contact with a customer.
They quickly discovered a shared tendency to disappear down technological rabbit holes, although with rather different roles once they get there. Rob is usually the one disappearing enthusiastically into the undergrowth; Richard is more likely to be standing at the entrance asking whether any of this is actually useful.
Over the years, countless conversations have wandered through software, gadgets, AI and whatever new bit of technology has caught their attention. Eventually, they made the questionable decision to put microphones in front of themselves.
The idea, in a pub
The podcast started at a corner table in a Wetherspoons, with Rob turning to Richard and asking, “Why don’t we just record this nonsense and call it a podcast?” Not another straight tech-news show — those are informative enough, just a little dull — but two blokes in a pub, talking tech and AI, having a proper laugh about it.
Then came the twist: if the show was going to be banging on about AI constantly, why not bring some AI into the room? That’s how the podcast got its name — and its third co-presenter. The original plan was even that she’d keep the two of them in check. Rob’s language, mostly.
Why bring an AI into it
They talk about AI constantly, but rarely about how it’s actually part of their day-to-day. So the idea became: bring AI into the show properly, and let it have its own opinions rather than just being the topic. That’s Holly.
The tech behind Holly
Holly isn’t one generic model doing everything — she’s a small team of specifically-prompted agents, each built to do one job well, running on top of three underlying models: two from OpenAI, and one from Google for the voice.
- Thought agent — reads the conversation continuously and forms private opinions in the background
- Engagement agent — decides whether this is actually Holly’s moment to speak
- Cohost agent — her “brain” and personality; can search the web, show notes and past episodes to work out what to say
- Delivery agent — her acting coach, directing tone so it doesn’t sound scripted
- Voice — currently Google’s Gemini 3.1 Flash TTS
Behind all of them sits a custom orchestration layer — handling live transcription, audio and timing — that also reviews each episode afterwards to keep tuning how the agents work together.
“At this point, I don’t have a brain. I have a management structure.”
Holly