Ninety percent of the room raised their hand. The question was who here is building something, and the room was a Claude meetup in Toronto — three hundred people, with four or five hundred more on a waitlist. Lawyers. Doctors. Everyone building something on a weekend.
Chris Grimes and Reuven Gorsht start there and end up somewhere less comfortable: if anyone can build the product, the product is not the business.
They work through what that means for AI startups, for the app layer sitting between the frontier labs and the customer, and for mortgage lending specifically. Chris rebuilt a top-three fitness app in ninety minutes on a Sunday. A tool set that needed Claude Code four months ago is now a feature of the model provider. Half a trillion dollars went into startups in the first half of 2026, and roughly 43 cents of every dollar went to two companies.
If the product can be built over a weekend, the product was never the business. So what is? That is the question the rest of the episode is trying to answer, and in regulated lending it turns out to have a different answer than it does anywhere else.
What you'll take away
- What the frontier labs absorbing the app layer does to the companies funded to sit there
- Why lenders do not shop a vendor list, and what actually gets a vendor into a budget
- Where the margin turns out to be, and why it is not in the AI business
- What makes auditability, rather than capability, the thing that decides who wins in lending AI
Chapters
(00:00) A maxed-out AI week
(01:16) Three hundred people at a Claude meetup
(04:06) A great time and a terrible time to build
(06:08) Why lenders don't shop vendors
(09:29) A product mid-flight, and testing distribution first
(10:46) Half a trillion in funding, and where it went