The hard part nobody talks about.

The way he put it was to imagine somebody else inherited your company and was asked to make it survive from here, with no emotional connection to the way it was built. What would they actually keep? When you run that question honestly, he said, you often find that the list is shorter than you want it to be. Brand, UI, product goals, pricing, differentiators, all of it potentially on the table. The only thing that probably survives intact is the data structure, and maybe not even that.

That's a confronting exercise, but it's the right one. His framing was the Ship of Theseus: if you change the brand, the UI, the product, the pricing, and the goals, what exactly are you preserving, and is that thing still worth preserving?

On the question of which software categories survive the next decade, his view was clear. Point solutions are in trouble. If you're the second or third biggest player in a workflow and someone owns the rest of that workflow, they'll just build your thing. The old strategy of entering as a wedge, expanding to a suite, and then to a platform will fade. What survives is whoever owns a strong system of record, something where a company genuinely stores critical canonical data. His examples were Salesforce and Workday. They might look slow from the outside, but they're the things businesses are least likely to rip out.

The question nobody asks him, he said, is what if it all goes right. The press has gotten very good at cataloguing the risks of AI, the ethics concerns, the environmental costs, the potential for misuse. Nobody asks what it would mean to give a basic education to four billion people who don't currently have access to one. Nobody asks what it would mean for healthcare. The upside, in his view, is potentially the largest thing humanity has ever seen, and it isn't getting the attention it deserves.

One of the more practical moments in the conversation came when I told him about building an event tech platform over a Sunday morning with Claude Code, three hours, 350,000 tokens, and something I could demo to the team. He recognised the pattern immediately. You can get to 40% very quickly. The other 40% will take you three years and sixty people, and you won't know that from the first three hours. His point wasn't that building is a waste of time. It was that you have to know whether you're doing it because it's your strategy, or because you had a quick win and extrapolated from it.

That distinction matters. Building because you're intentional about it is one thing. Building because Claude Code made it feel easy is a different thing entirely.

The full conversation is on The AI Revolution Show. Subscribe on Apple Podcasts: