As AI Commoditizes Agency Work, Deep Vertical Expertise Becomes the Defensible Edge

Nine in ten US marketing agencies now use generative AI, and half have handed marketing execution to agentic AI, according to Forrester's 2026 study with the 4As. When a prospect can spin up an agent to write the copy, build the campaign, and pull the report in a few clicks, the work an agency used to bill for is suddenly something the client can do alone. That puts pressure on the agencies whose value has always been tied to execution.

Corey Quinn's argument starts there. AI is very good at execution, he says, which is exactly why he thinks execution is about to stop being worth much. Quinn spent years as CMO of Scorpion, where the agency grew from around $20 million to $200 million by serving a couple of narrow verticals, attorneys and home-service businesses, and he later wrote a book about the approach, Anyone Not Everyone. What worries him now is the agency that does a bit of everything for everyone. Once the copy and the campaigns are a few clicks away, the generalist has handed over the only thing it was selling.

Quinn's bet is that deep specialization produces a kind of judgment a machine can't easily reproduce: the edge cases and context that come from doing one kind of work for one kind of client a thousand times. He quotes Bruce Lee to land it: fear the man who practiced one kick ten thousand times, not the one who practiced ten thousand kicks once. "The agencies that will win won't be the ones with the most services," he says. "They'll be the ones who go the deepest in one place."

You can't automate chaos

The second problem shows up when a service business tries to scale without adding headcount. AI works best when the work is narrow enough to systematize. A generalist has too many one-off workflows to train an agent against.

He gave a specific case. One of his clients, a digital agency, builds websites and runs SEO and Google campaigns for hospitals and healthcare organizations. Those clients kept telling the agency the same thing: the phone's ringing, but revenue isn't moving. So the agency pulled the actual phone calls and listened. The marketing was working. The intake wasn't. People were calling in, getting a clumsy experience from whoever answered, and going elsewhere. Because the agency worked only in healthcare, it could build an AI intake agent that understood the patient, the problems being solved, and the specific kind of empathy a medical call needs, which is nothing like a plumbing call or an airline call. The client adopted it immediately, and it compounded the results the marketing was already producing. "They would build a general intake agent that would be bland and would probably be off the mark," Quinn said of what a non-specialized agency would have shipped instead.

When the founder stops executing

Once agents handle more of the delivery, Quinn thinks the founder's job changes too. He'd rather see that time go into intellectual property and deals tied to results instead of hours. AI makes it more practical to shift that time away from delivery. And selling the outcome rather than the hours lines the agency's incentives up with the client's.

His own favorite use is a version of exactly that. Quinn teaches a consultative sales process to founder-led agencies who want to stop being the only person who can sell. The usual failure is hiring a salesperson and expecting them to sell like the founder, which they can't, because they don't have the founder's tools. So Quinn wires the call recordings, from Fathom or Fireflies, into an AI layer that scores each call against the playbook: did the rep listen actively, did they ask for the sale, did they follow the choreography. That used to take a human sitting in on the call with a notepad. Now his clients get a detailed readout whenever they want it, and he says they'd rather work through the bot than book time with him. "They love it more than talking to me," he said.

Quinn takes a pragmatic approach to build versus buy. A firm that wants to move fast can plug in a third-party tool and get what it can out of it, then build its own version only when the off-the-shelf one can't do the specific thing it needs. For most startups, the founder's time is better spent with clients than learning to build an AI product from scratch. He's also watched a particular trap play out: a CEO builds a website audit tool or a calculator, decides it might be the next Salesforce, and wants to commercialize it, missing that everyone else is building the same scanners and calculators. The widget was never the advantage. The expertise that made it worth building was.