Everyone Became an AI Agency Last Year
Same people, same skills, new page on the website — why most of it didn’t work, and what does.
Sometime in the last eighteen months, almost every agency you’ve worked with became an AI agency. The marketing shop. The web studio. The SEO firm. The IT company that set up your servers. Same office. Same people. Same playbook as the year before — with one new page on the website, and the letters “AI” on it.
I get the pull. There’s a gold rush on. And bolting “AI” onto your homepage is far easier than learning to build it. So that’s where most of the effort went — the easy stuff. A bot to answer the simple questions. A voice agent. A clever prompt. All of it riding the same wave that, for a while, lifted any boat with “AI” on the side.
Then the owners who paid for it started asking the real question. Did it move the needle? Did it cut costs? Did it free up my people? Did my apps and spreadsheets finally start talking to each other? Mostly, no. And you’re not imagining it. MIT studied more than 300 companies using AI. About 95% saw no real impact on profit. These are firms with budgets and teams most of us don’t have — and almost nothing landed.
Here’s the reason, and it’s the part nobody wants to say out loud. The value was never in the tool. McKinsey looked at twenty-five things that might explain why some companies profit from AI and others don’t. The one that mattered most? Whether they redesigned how the work flows. Their words for it are blunt: the winners don’t bolt AI on. They rebuild the process.
And you can’t rebuild a process you don’t understand. In a real business, the work is rarely one clean feature. It’s a knot, with years of history around it. You’ve bought a dozen apps, each fixing one thing and adding two new seams. The business still runs on a spreadsheet someone guards like a state secret. Your systems don’t talk, so your people are the glue — copying numbers from one screen to the next, all day. The real job is tying up the loose ends of the legacy systems you already run. Weaving them into one thing that holds. That takes someone close enough to see how the parts connect — the workarounds nobody wrote down, the thing Jane does every Tuesday that the whole place quietly runs on. The code was never the hard part. Knowing what to build is.
Which is why the “AI” label on a website tells you almost nothing. Gartner even has a name for the past year: “agent washing.” It means slapping “agentic AI” on an old chatbot, with nothing new underneath. So the question isn’t whether a firm has AI on its homepage. It isn’t whether you pay monthly or by the project. It’s one thing. Does the person doing the work get close enough to understand your business and rebuild it? Or is your company just one more, getting the same template as everyone else? That’s true of an agency, a strategist, or an in-house hire alike. How deep they go and how close they get — that’s what matters. The rest is packaging.
The gold rush will end the way they all do. A lot of the “AI” will quietly come off the websites. But the businesses that get real value from it won’t be the ones who bought the most. They’ll be the ones who understood their own operation well enough to rebuild it. AI was never something you buy off a shelf. It’s work. And the work starts with understanding what you do.
— Sandeep Dhall. I write about how owners of real businesses use AI and the other levers that grow, turn around, and build lasting value — the ones with real money on the line. Co-founder of Bridge.Digital, solving real business problems with AI and custom software. Founder of smbcapitalpartners.com, where we buy, build, and back small businesses worth keeping.