Moats in the age of AI: what still defends a startup
When anyone can call the same model, the model is not the moat. Proprietary data, workflow ownership and distribution are.

In this article
Insight prepared with AI assistance and reviewed by the Winter Kpital team.
The commoditization problem
Foundation models improve and get cheaper every few months. Sequoia and a16z have both argued that thin wrappers on top of a model are easy to copy. If your product is a prompt, your competitor is one weekend away.
What still compounds
Durable advantages look familiar: proprietary data that improves the product with use, deep integration into a customer's daily workflow, switching costs, trust in regulated industries and distribution the incumbents cannot easily replicate. Ben Thompson's Stratechery framing of aggregation and Hamilton Helmer's 7 Powers remain useful lenses.
Speed is a strategy, not a moat
Shipping fast buys learning, and learning buys data and relationships. But speed alone erodes. The goal is to convert speed into one of the compounding advantages above.
How to test yours
Ask three questions: if a well-funded competitor used the same model tomorrow, what would they lack? Does the product get better for each customer as more customers use it? Would a customer lose something real by leaving? You can run this analysis with our private moat evaluator.
Sources
Sequoia Capital. Andreessen Horowitz. Stratechery by Ben Thompson. Hamilton Helmer, 7 Powers.
