There is a quiet difference between agencies that say they "use AI" and agencies that have actually built something with it. The first group is renting. The second group is operating. Most of the market is in the first group and they do not know it.
Andrej Karpathy, co-founder of OpenAI, called this distinction software 2.0. The old way of building software was writing rules. The new way is training a system on data and letting the patterns emerge. Marketing has the same split, and almost nobody is treating it that way.
The rental model
The rental model looks like this. Agency hires a copywriter. Copywriter opens ChatGPT. Copywriter pastes a brief, gets ten variants, picks the one that reads cleanest, ships it. The model did the work. The agency took the markup. The buyer got a slightly faster version of the same generic creative every other agency is shipping.
Nothing about that workflow gets smarter. The next campaign starts from the same blank prompt. The lesson from the last winner does not carry forward. The model is doing exactly what it would do for the agency down the street that pays the same $20 a month for the API.
When the same tool is available to every agency on earth, the tool is not the asset. What you build on top of it is.
The operator model
The operator model looks different. The agency is not asking the model questions. The agency is feeding the model context, then asking it to do work against that context.
What does context look like? It looks like this:
- The buyer's ICP, scored, with the disqualifiers spelled out
- The brand voice, cloned from existing assets, with examples of what does and does not sound like the brand
- The offer, documented at the level of detail it would take to onboard a new senior employee
- Every ad ever shipped in this account, tagged with what worked, what did not, and what we learned
- Every ad ever shipped in every other account in the same niche, with the same tags
Now when the model writes copy, it is not generating from scratch. It is generating against a knowledge base that knows the buyer better than the buyer's last marketing agency did.
Why this is software 2.0 in an ad account
The classic software 2.0 idea is that the system learns from data, not from code. The agency version of that is: the system learns from every ad you ship, and every ad we have ever shipped on behalf of someone like you, and every result we have ever logged.
Three things have to be true for this to actually work:
- The data has to be structured. Every creative shipped is logged with its script, its audience, its cost-per-result, the ICP it landed for. Without that structure, you have a folder of ads, not a learning system.
- The feedback loop has to close. The model has to see what shipped, what scaled, and what got killed, in real time, so the next variant ships smarter than the last.
- Someone has to steer. This is the part most "AI marketing" pitches skip. The model can ship variants. The model cannot decide what gets approved. That is a senior marketer's job, every time.
What this looks like for the buyer
If you are the operator on the other side, here is what changes when an agency moves from rental to operator:
The first creative that ships in your account is not generic. It is built against the audiences your existing customers actually look like. The pages it lands on are written in your voice. The follow-up is sequenced against the objections your sales team is already used to handling.
By month three, the system is faster than the people who built it. By month six, it is shipping versions of campaigns that work because they have already worked, in your account or in an adjacent one we have already operated. None of that compounds when you rent. All of it compounds when you build.
Most agencies are renting a tool. We are operating a system that compounds on every ad we ship and every metric we track inside your account.
The market is going to split
In two years, the agencies still operating on the rental model will be priced like utilities. The agencies operating on the system model will be priced like operators. The gap is already opening.
The reason almost nobody is on the right side of that gap is because building the system is harder than buying the seat. It takes a senior marketer who knows how the buyer thinks. It takes infrastructure that logs every ship, every result, every kill. It takes the discipline to feed the system back to itself, every week, even when no one is watching.
That is the work we are doing. That is why we built this.