AI will expose the agencies that were phoning it in

Every few months someone asks us, with varying degrees of tact, whether AI is coming for agencies. The honest answer is yes, but not the way the headlines frame it. It isn’t coming for agencies. It’s coming for a specific kind of agency work, and whether that’s terrifying or liberating depends entirely on what you’ve been selling.

The floor just moved

Here’s what actually changed. For twenty years there was a floor under agency pricing made of pure execution: the hours it took to produce a competent landing page, a serviceable logo exploration, a readable blog post, a working template. That work wasn’t brilliant, but it took time, and time was billable. A lot of the industry lived comfortably on that floor.

The floor now belongs to the machines.

Competent, generic, grammatically flawless, technically functional output is approaching free. Not good output. Average output. And it turns out an uncomfortable amount of what got invoiced as strategy and craft was, held up to the light, average output with a nice deck around it.

That’s the exposure. AI doesn’t threaten teams doing original thinking, because it can’t do their job. It threatens business models built on reselling effort, because effort just got cheap. The template with your logo on it, the recycled strategy document, the blog packages written by no one for no one: none of that got worse this year. Its true price just became visible.

The ceiling didn’t

Now the part the hype skips.

The ceiling, the thing that separates work that moves a business from work that merely exists, hasn’t budged, because the ceiling was never made of execution. It’s made of judgment. Knowing which of two hundred plausible directions is right for this client, this market, this moment. What to leave out. Recognizing that the brief is wrong, which it often is, and having the standing to say so out loud.

AI generates options at a speed that’s genuinely astonishing. What it can’t do is want something. It has no stake, no taste, and no capacity for embarrassment, and embarrassment, honestly, is load-bearing in creative work. The feeling that says “this is fine and we can’t ship it” is not in the training data.

We use these tools daily, and we like them. They’re excellent at the eighty percent that was never the point: boilerplate, scaffolding, first passes, alternate versions, the tedious middle of things. What they hand you is a floor to stand on, not a place to stop. The gap between the machine’s draft and something worth a client’s money is now, precisely, the job.

There’s an honest, unresolved question inside all this, and we’d rather name it than pretend otherwise. Taste is trained by repetition. Designers develop an eye by producing a thousand mediocre layouts, writers develop an ear by writing badly at length, and a lot of those reps used to happen on exactly the routine work the machines now absorb. If the junior work disappears, the industry has to get deliberate about how people become senior, because judgment was never innate. It was accumulated. That’s a problem worth solving on purpose rather than discovering by accident in ten years.

What this means if you hire agencies

When average work is free, paying for average work becomes the one truly indefensible budget line. That’s actually good news for anyone buying creative services, because for years the hardest part of hiring an agency was telling the difference between substance and production value. The tools just did you a favor: production value stopped being evidence of anything. So the questions worth asking your partners get sharper.

Ask for the reasoning, not just the deliverable. Anyone can show you a polished thing now. Fewer can explain why this thing, why for you, why not the other forty versions. If the answer to “why” is a shrug wrapped in jargon, you’ve learned what you needed to.

Then ask how they use AI, and be suspicious of both extreme answers. “We never touch it” means paying artisan prices for typing. “It does everything” means the judgment you’re actually buying is nowhere in the building. The credible answer lives in the middle and sounds specific.

And watch the estimates. Work that’s mostly execution should be getting faster and cheaper, honestly and visibly. If the pricing hasn’t moved but the deliverables have acquired that faint, weightless, nobody-was-home quality, the savings went somewhere, and it wasn’t to you.

Neither the funeral nor the fireworks

We don’t buy the doom, and we don’t buy the revolution either. Every tool that made production easier, from desktop publishing to WordPress to the site builders, was supposed to end this industry. Each one ended only the version of it that was coasting, and then the work regrouped around whatever still required a human with a point of view. It’ll happen again. Faster, this time.

What’s genuinely new is the honesty of it. For years, effort and value could blur together, because effort was expensive and hard to inspect. Now effort is cheap, the blur is gone, and everyone in this business gets to find out which of the two they were actually selling.

We’ve thought about that question a lot, fourteen years in. We’re comfortable with our answer.

The machines make average free. Somebody still has to be worth paying for.