The agency model does not sell tools. It acts as the middle-man, selling and running the tool, for as long as someone keeps paying.
The agency model predates AI by a century — advertising agencies formalised it, historically taking a commission on the media they placed on a client’s behalf rather than charging a flat fee for their own labour. That structure creates a specific incentive problem economists call agency theory: the agency’s interests (more spend, more billable work) aren’t automatically aligned with the client’s interests (better results, lower cost). Spake, D’Souza, Crutchfield and Morgan’s 1999 study applies agency theory directly to advertising-agency compensation, showing how different fee structures (commission vs. flat fee vs. performance-based) change what the agency is actually incentivised to optimise for.
Applied to AI, the same logic holds: understanding the basics of how one can use AI and keep selling it. The AI itself is often not the moat — anyone can access the same underlying models. What’s actually being sold, agency-style, is the ongoing service of running and maintaining it on the client’s behalf.
Source: Spake, D.F., D’Souza, G., Crutchfield, T.N. & Morgan, R.M. (1999), “Advertising Agency Compensation: An Agency Theory Explanation,” Journal of Advertising, 28(3), 53–72.