Product-market fit wins in AI
Kavin Stewart, Partner at Tribe Capital, on Reddit's 10x opportunity
The key investing problem in early AI is not whether the demo looks magical, it is whether the product keeps enough control over demand, pricing, and distribution to become a repeatable business. That is why the focus shifts to product market fit and unit economics instead of broad narratives about model moats. In 2023, even strong products could still be compressed by open source models, new model releases, or platform owners bundling similar features into larger products.
-
The art generation example captures a real business model trap. If image quality keeps improving across open and closed models, the hard part is not generating images, it is owning a workflow people pay for every week, like ad creative production, design review, or ecommerce listing creation.
-
This matches Tribe's underwriting style. The firm built its process around normalizing company data and comparing businesses on concrete operating signals. That framework works better when the technology stack is moving too fast for long range category predictions to be reliable.
-
The same interview points to why platform risk felt manageable but application risk did not. Multiple model providers were already closing gaps, Google was integrating Bard with search and other products, and Anthropic had expanded Claude to 100K tokens. That made model capabilities more available, but made it harder for any single app idea to stay unique.
The next phase favors AI companies that lock themselves into daily workflows, proprietary data, and clear budgets instead of relying on novelty. As foundation models converge and distribution shifts toward large platforms, the winners will look less like thin wrappers and more like software products that save time, drive revenue, or replace headcount in a way customers can measure.