In June 2025, I went on 21 Hats and told Loren Feldman that the general purpose agency was doomed. My logic was simple: if you charge $5,000 to $10,000 a month for work AI can now approximate, someone will eventually go get the AI to do it for them. It was around then that I first realized how transformational AI would be.
Last week, Loren brought two of the owners from that conversation back: Jaci Russo, who runs a branding agency, and Sarah Segal, who runs a PR firm. They looked at my prediction and said, more or less, that it hasn't happened. Sarah says AI has helped her. The models lean on credible news coverage, so clients are coming back because, as she put it, you won't show up in Gemini, Claude, or ChatGPT without third-party validation. Jaci says AI handles the routine work and frees her team for events, strategy, and relationships.
They’re right. I also think they're describing the last fifteen months, not the next three years.
The Innovator’s Dilemma
In his classic book, Clayton Christensen describe how startups overcome incumbents. When the startup first comes to market, it’s services and/products are substandard. The incumbent looks at the output and laughs. Most customers don’t switch. Only the lowest-end most cost-conscious do. The startup tends to peel off the lowest-value and lowest-margin work.
But that revenue helps the startup survive and it works away at its product and services. Over time it eats more and more of the spend from the bottom up. It maintains the same cost or innovation advantage as it goes.
Eventually the incumbent isn’t laughing and tries to catch up when it’s too late. This is Netflix with Blockbuster or Borders with Amazon or PCs with mini computers and then Compaq with IBM and then Dell with Compaq.
My take is that this same process will play out in professional services. Of course AI isn’t replacing someone with 30 years of domain experience today but it is and will continue to start carving out those “lower-value” services.
Unfortunately those low-value services support a lot of the professional services revenue. Agencies have an analogous problem to newspapers.
Newspapers Sold Classifieds Not Journalism
30 years ago when newspapers were at their peak, nobody bought them for the classified ads but those ads were the business model. They paid for the reporters, the editors and the foreign bureaus. Then job listings, apartments, porn sites, and used cars moved to websites that each did one thing well. That gutted the newspaper model.
Professional services firms have their own classifieds: the junior work at the bottom of the pyramid that pays for the partners, the brand and the training. AI-native companies will come after that layer first.
As those companies work their way up the value stack, they’ll kick out the traditional legs of the proverbial professional services stool one by one. Ad design, ad targeting, brand, logos, websites, copy, social media creation, etc. etc. All that will be left is human judgement, taste, and relationships.
Those things are important but it will require a complete transformation of most agencies’ business models to keep them as the rest is eaten. (I’ll write next week about why many of these services will erode as well.)
The Most Important Thing is Timing
My expectation is that few will actually make the transition. It requires new skills, disrupting employees and teams that don’t want to be disrupted, and at first it’ll look like the wrong move.
It also won’t feel urgent until it does. Pew Research Center showed that newspaper ad revenue was still going up in 2005. That’s almost a full decade into the internet revolution. But then it just cratered.
Newspaper ad revenue went from $50b to less than $10b in 10-15 years.
Retail is another example. It moved differently than newspapers. Online sales went from under 1% of US retail in 2000 to 17% in mid-2026, about a point a year.
There was no single cliff. Instead there was a long list of bankruptcies: Circuit City in 2009, Borders in 2011, Toys "R" Us in 2017, Sears in 2018. Amazon's revenue passed Sears' in 2011, when Amazon was seventeen years old.
The stores that survived were discounters or the highest end. The middle got devastated.
Why it took so long, then and now
Why did these declines take so long?
Entrepreneurs had to create new services and companies. Clients needed broadband. Web developers needed new languages, frameworks, and services to build websites. The iPhone had to revolutionize mobile.
Once those declines started they were relentless, but the picture was clear early on. In fact, many of the ideas that would cause that decline were already present in 1999 and 2000. Looking back from 2014, Marc Andreessen, the venture capitalist and founder of Netscape, said: "I basically think all the ideas of the '90s that everybody had about how this stuff was going to work, I think they were all right, they were all correct. I think they were just early."
Pets.com became Chewy.com. Webvan became Amazon. Yahoo became Google. Kozmo.com became DoorDash and Uber Eats. Geocities became Facebook.
AI is following the same script. I called the direction right on 21 Hats but it will take time and it will seem like we’ve reached safe shores before the bottom falls out. It’s partly why I stopped just writing about it and started doing it, founding Rustproof.ai last summer.
As Mike Campbell says to Bill Gorton in Hemingway’s The Son Also Rises:
"How did you go bankrupt?" Bill asked.
"Two ways," Mike said. "Gradually and then suddenly."