2 builders who never spoke to each other

Say which market you are in, not that you use AI

Unless you train the models, you are competing in the category you were always in and it now has tougher entrants, including incumbents who own both the installed base and the data to train on. Plan where the proprietary data comes from and treat getting to buyers as the whole race, since the incumbent already starts with the innovation.

The method

1. 01

   Do not assume incumbents will ignore AI the way they ignored earlier technologies: they are spending heavily on it, so you are competing head on with companies selling into an installed base.

2. 02

   Expect model capability to keep getting cheaper while training data gets more expensive, because the companies sitting on the data have worked out what it is worth.

3. 03

   Frame the opportunity as how this technology reshapes one existing market and where you sit inside that change, rather than as a share of a trillion-dollar total.

4. 04

   Plan how you will get proprietary data before you plan the product, because the customers an incumbent already has are also the data they can train on.

5. 05

   Stop describing yourself as being in the AI market: unless you are building models, you are in the market you were always in, and it now has tougher competition.

What it returned

• Cohen contrasts this with the transistor radio pattern, where the incumbent dismissed the cheaper, worse product until it improved.

• He points out AI can enlarge an existing market two ways: current buyers pay more for a much better version, and buyers who previously refused the category on quality grounds now enter it.

• Intercom builds on hundreds of millions of customer conversations, Google on owning the YouTube library, and GitHub trained Copilot on its own repositories including the human explanations attached to changes.

• A chatbot competes in the chatbot market and a search tool in the search tool market; the underlying customer problem, such as answering customers around the clock in any language, has not changed.

Sources

• AI startups require new strategies: This time it’s actually differentJason Cohen · longform.asmartbear.com · Disruption Theory does not apply to AI · 2024-03-03
• AI Employees: Named Personas, Cautionary Tale, Outcome Pricing - Trends.vcDru Riley · trends.vc · ☁️ Opportunities · 2026-05-21
• Personal AI Agents: Skill Marketplaces, Compliance Wedge, Open Weights - Trends.vcDru Riley · trends.vc · ☁️ Opportunities · 2026-05-28
• Privacy as a Service: Indie Opportunities, Product Bundles, Audit Moats - Trends.vcDru Riley · trends.vc · ☁️ Opportunities · 2026-04-23
• Software Development Services: Attracting New Clients, AI Apps, No-Code Tools - Trends.vcDru Riley · trends.vc · 💡 Solution · 2023-09-27