Anatomy of an AI startup
A cross-section of a typical AI company, from interface to bedrock, ranked by how hard each layer would be to copy.
the whole body hangs from one rope
trying to answer →Which parts of an AI startup are actually defensible?What is a moat in an AI world?
hangs from someone else's APIinterface (thin)workflow = the muscleintegrations: the gripVesalius’s plate is a useful picture of a lot of AI companies: an impressive, detailed body hanging from a single rope it doesn’t own. This analysis takes the body apart, layer by layer, and asks of each: could a well-funded competitor with the same model copy this within a month?
Above the erosion line
The interface, the prompts and the choice of model are what users see and investors demo. They are also what a competitor can reproduce fastest. Prompts leak through outputs; interfaces are screenshots; the model is rented from the same three suppliers everyone else uses.
Below it
Things get harder to copy as you go down. Workflow embedding and integrations are slow to earn: permissions, security reviews, the habits of a team. Proprietary data only counts if it compounds, meaning the product gets better because it is used. At the bottom are relationships, distribution and accountability. None of these are software, which is exactly why software can’t copy them.
The uncomfortable implication
The layers that get the most engineering attention are the least defensible, and the most defensible layers look like sales, support and compliance. An AI startup that wants a moat may need to be less of a technology company than it thinks it is.