The first of three essays arguing that a law firm's real exposure is not AI, it is renting the layer its differentiation lives in.
Written a year before the sovereign stack existed as a product, this piece asks the question the legal-AI market was avoiding: if most of what firms are buying is a wrapper around somebody else's model, what exactly has the firm acquired?
The argument is about dependency rather than capability. The labs underneath those wrappers lose enormous sums monthly, and they have both the technology and the incentive to move up the stack into the vertical their customers occupy. A vendor's differentiation, built from custom filtering and workflow constraints, is thin protection against the party that controls the model.
The alternative it proposes is the one we now build for a living: open-weight models on the firm's own infrastructure, at an infrastructure cost a mid-sized firm can carry, with the firm as the software provider rather than the software's customer.
This is our summary, not the article. The full text stays with the original publisher. Read it there.