The brand on the install.

The major AI model makers — the foundation-model companies whose product everyone is now building on top of — sell at enterprise scale through forward-deploy teams and alliance partners. They fly in. They configure. They put their name on every install. The case study has the model maker's name on it. The annual report has the model maker's name on it. The onstage testimonial has the model maker's name on it. The customer remembers the model maker. The model maker compounds.

That's how their distribution works. By design. The math of a six-month forward-deploy engagement only works if each customer becomes a referenceable case study under the model maker's brand — that's how the next enterprise sale gets cheaper. The model maker is doing honest distribution. It has to be. That's the only way the unit economics work.

When the model maker installs, the customer remembers the model maker. When the advisor installs, the customer remembers the advisor. Same compute underneath. Different relationship.

There's a layer of work the model maker structurally cannot do. A model maker can ship a capability — a model trained on the world, available through an API, behaving the same way for every customer who calls it. They cannot ship a configuration of that capability for one specific business. They have one product; their customers number in the millions. The model maker doesn't know your pricing, doesn't know your customers, doesn't know what your team said yes to last week. Their business is the product. The configuration work is somebody else's job.

That somebody is usually already in the customer's life.

The local advisor — the agency principal who has run the customer's marketing for six years, the consultant whose name is already in the customer's phone, the operator whose people the customer has already met — installs the workforce around how this specific business already runs. They configure the agents to know the customer's pricing, customer base, competitive landscape, the way the team actually works. They answer when something breaks. They were the relationship before the workforce arrived. They will be the relationship after this generation of models gets replaced.

The model maker can ship a capability. They cannot ship a configuration of that capability for your specific business. That's somebody else's work.

This is the layer the customer's memory holds. Whichever model runs underneath gets swapped on a cadence the customer never has to read. The substrate underneath gets versioned. New vendors arrive. Old vendors release. The customer doesn't track any of it — that's a working substrate's job description, to disappear. What the customer tracks is who showed up. Who configured. Who answered the call when something went sideways. The brand on the install is the part the customer remembers, and that's the part that compounds into next year's renewal, next year's referral, the customer's CFO's first call when a peer asks how they figured this out.

The brand on the install is what compounds for whoever installed it. For the model maker, that's the enterprise customers — and that's healthy, because that's the work the model maker is actually doing for them. For the advisor, that's the customer relationship that just expanded into a new layer of work — and that's also healthy, because that's the work the advisor is actually doing. Two different brands on two different layers. Both honest. Two beneficiaries.

The brand on the install is what compounds for whoever installed it. For the model maker, that's the enterprise. For the advisor, that's the relationship that just got bigger.

A customer who went direct to the model maker would get a capability. A customer who works with the advisor gets a workforce configured around their specific business by someone who already knew their business. Same compute underneath. Different relationship. Different fit. The advisor isn't a middleman on top of the model — the model maker structurally couldn't deliver what the advisor delivers. The advisor is the layer that makes the capability fit the customer.

The customer ends up doubly free, too. Not just free of the model maker's brand on the relationship — also free of the model maker's choice of model. The substrate underneath the workforce picks the right model for each task; when a smaller, faster model lands next quarter that does the simple work better, the substrate routes to it without anyone telling the customer. The bill goes where it goes; the work decides. The advisor isn't locked to the model maker's brand or roadmap. The customer isn't locked to either. Two layers of freedom, accumulating on the side of the relationship that already mattered.

The math of who reaches whom is the rest of the story. Model makers reach the customers they can afford to send forward-deploy teams to. Below the floor where a six-month enterprise engagement makes financial sense, the model maker cannot reach — not because they don't want the customer, but because the unit economics of their distribution don't survive the trip down to SMB scale. Below that floor, the customer is reachable only through people they already trust. The local advisor. The agency owner. The consultant whose name is already in the customer's phone.

Models churn on a cadence the customer never has to read. The brand on the install does not.

The shape of distribution at SMB scale was always the same one — the local relationship was the asset. What changed is that the local relationship can now carry a workforce under its own brand without the unit economics blowing up.

Models churn. Installs stay. The model maker gets the enterprise. The advisor gets the relationship. The customer gets the work — configured around their business, by someone who knew their business.