WildEdge Worx cofounder and CEO Claire Ackers built an AI ‘dream team’ – a virtual board that never eats, sleeps, or schedules an emergency meeting to discuss homepage fonts on a Friday afternoon.
It does, however, challenge her ideas, stress test her plans, and produce content in her voice.
At Greenhouse’s recent AI Show ‘n’ Tell, held at Leeds Beckett’s innovation space, she showed a room full of AI-curious delegates how to do the same.
The self-styled ‘Business Leadership Growth Engineer’ recognised early on that scaling would be a challenge. WildEdge Worx takes business leaders outdoors for breakthrough days and retreats – and with only her and husband Anthony (cofounder and COO) as permanent team members, it runs almost entirely on the two of them.
And so, Claire turned to ChatGPT in an attempt to speed up content creation and strategy planning. Like most of us, she tore her hair out when it spluttered out em dashes, stacked sentences, and a raft of other stylistic habits that sounded nothing like her.
Teaching the machine to sound human
Claire took a different approach. Instead of repeatedly fixing the LLM’s output, she changed something that she could control: the input.
She uploaded transcripts of 162 podcast episodes of herself speaking and began conversing with the LLM using ChatGPT’s Voice mode while out walking in her local area, which gave the model her cadence, rhythm and vocabulary. She then shared her grammatical preferences as a rule set, banning the aforementioned telltale patterns that make AI-generated content so easy to spot.
There was finally a shift. The model was starting to sound like Claire.
Building an AI dream team
If Claire could teach the model to replicate her voice, she figured she could also teach it to replicate expertise.
She built a team of AI personas on the same principle: a launch director, a finance director, a social media director, and a PR consultant, each grounded in the thinking of a real person rather than what a language model thinks that role should sound like.
Her finance director, for example, was built by going for a walk with a real FD, recording the conversation, and feeding it into the model. So when it responds, it’s drawing from actual knowledge, not OpenAI’s training data.
As Claire put it, ‘you can’t ChatGPT your way up a mountain’.
The board meeting
The live demo she ran showed what this looks like in practice. One prompt called a virtual board meeting, tasking the personas with drafting a two-week marketing strategy for a product launch. A second turned that strategy into a printable action checklist, with the FD adding projected revenue against each action.
If you want the full step-by-step of how she built the personas operation, Claire is running a free dedicated virtual session on exactly this. It takes place on Thursday 26 February at 9:15. To attend, drop her a message on LinkedIn and she’ll share the Zoom link with you.
For updates on future AI Show ‘n’ Tell events, follow Greenhouse on Eventbrite.
