Every week we hear some version of the same prediction: AI is going to compress the agency business. Faster outputs, fewer hours billed, automated reporting, automated content. The headline forecasts have agencies of three doing the work of agencies of thirty by 2027. We don't disagree with the math. What we disagree with is the conclusion.

The agencies that disappear in the next five years won't be the ones who failed to adopt AI. They'll be the ones who treated their work as transactional in the first place. AI is genuinely good at the transactional layer of marketing, pulling reports, drafting copy, generating creative variations, summarizing research. That's exactly the layer that was never the moat. The moat was always the relationship.

What AI can't do, even now

Spend a week running a marketing program with a sophisticated AI co-pilot, and you'll find the same gaps everywhere:

None of this is a knock on AI. It's an observation about what the value of an agency relationship actually is. Most clients aren't paying for the deck. They're paying to outsource a piece of judgement.

Most clients aren't paying for the deck.
They're paying to outsource a piece of judgement.

What we've watched work

Across our portfolio, the engagements that compound are characterized by four things, none of which AI replaces:

01Trust earned in the first 30 days

The agencies that retain clients long-term tend to deliver one un-asked-for, high-leverage observation in the first month. Not a deliverable; an insight. Something that says "we've been paying attention to your business, not just the brief." That moment of recognition is often what makes the relationship survive the eventual rough quarter.

02Customization beyond the platform reports

Generic platform recommendations are now free. AI will produce them on demand. What clients actually want is a strategy this account would benefit from, given this business model, this team's capacity, this moment in the category. That's a human judgement, informed by context the model doesn't have.

03Adaptability in real time

The most flattering metric for any agency relationship is how quickly the team can pivot when something breaks. A media platform shifts attribution rules; a creative concept tests poorly; a competitor launches an aggressive campaign on day 6. The agencies who pivot in 48 hours are the ones who survive these moments. AI helps execute the pivot. The decision to pivot is human.

04Partnership built on shared outcomes

The healthiest engagements we've seen are structured so the agency wins when the client wins, not on hours, not on retainers, not on outputs. That alignment doesn't come out of an SOW. It comes from the relationship.

What this means for how to grow

If you run a small agency and your AI strategy is "we're going to be faster than the competition," fine, that's necessary. But it's not sufficient. The faster you get with AI, the more your differentiation collapses to relationship quality, because everyone else is also getting faster.

The implication: the time you save with AI isn't a margin gain to pocket. It's bandwidth to spend deeper into client relationships. More on-site visits. More unprompted insights. More time spent thinking about the client's business.

That's the trade. AI compresses the time required to produce the work. The agencies who pocket that time as margin will get out-competed on price. The ones who reinvest it in relationship depth will compound.

The shape of the next 24 months

Here's what we expect to see across the small-agency landscape:

None of this is a defense of doing things the slow way. AI adoption is non-negotiable. But adoption is the price of staying in the game, not the path to winning it. The path to winning still runs through the same place it's always run: a senior person, paying genuine attention to a client's business, with the judgement to make the right call.