Next Gen Builders

The episode contains solid practical insights about AI deployment in risk-averse industries, trust mechanisms, and organizational innovation structures, but is diluted by extended biographical narratives, generic leadership advice, and repetitive points about change management that won't surprise experienced operators.

“if the model output says, I don't know what the answer is here, then the agent will go back to the human and say, I don't know what steps to take next, because we're not quite sure what it is”

“we've taken customers like Hitachi from days of creating contracts to getting them signed to now five minutes”

While the multi-model orchestration approach and probabilistic consensus for hallucination mitigation show some technical freshness, much of the positioning around AI in legal, trust in enterprise AI, and culture change repeats well-worn startup narratives without contrarian perspective or unexpected angles.

“you can start to train the models, um, to look at getting probabilistic consensus and answers. And if there isn't probabilistic consensus, then the AI can go back to the human and say, actually I don't know the answer”

“if a model comes out tomorrow, that's better than us, and then we just rip and replace it”

Eleanor Lightbody is a proven operator - she's built distribution and sales at Darktrace (acquired for $6B), scaled across multiple geographies, and is actively running a funded AI company serving enterprise customers at scale. However, she lacks deep domain expertise in law itself, which limits her ability to speak authoritatively on legal industry pain points beyond what she's learned on the job.

“I left just before the IPO and it just got bought by Thomas Bravo for over six um, billion”

“went to run the Middle east and Africa for a bit and then uh, looked after a global product that looked at securing national critical infrastructure”

Good use of named customer examples (Hitachi, Panasonic, AMD, DHL, Yokoga) and specific quantified outcomes (days to five minutes, seven to ten days to five to seven minutes, early payment discount recovery), but limited on technical specifics, product architecture detail, or market sizing data that would help operators replicate insights.

“we've taken customers like Hitachi from days of creating contracts to getting them signed to now five minutes”

“they would have taken like 10 days, that seven to 10 working days to answer any business questions. And now it takes them five minutes, um, five to seven minutes”

The host asks competent opening questions and some decent follow-ups, but rarely pushes back, challenges claims, or digs deeper into contradictions. Questions about SVB crisis and mentor relationships feel tangential rather than probing the core business model. The conversation remains predominantly in Eleanor's comfort zone of storytelling rather than rigorous testing of assumptions.

“Well, I mean, I think AI is really good at unstructured data reading and writing text, which would seem to make it really applicable to AI. When do you think about the start of luminance?”

“And do you think that as a leader you've learned, I mean there's a lot of pattern matching that happens there, but have you learned that through selling and doing discovery in deals and like how you see the world”