Commodity Drift Is the Real AI Risk for Agencies

The real threat of AI isn’t to your work. It’s to how clients perceive the value of that work.

An agency I work with lost a client last week. Not to a competitor. The client had started cranking out a blog post a day themselves, watched their traffic climb, and figured they didn’t need the agency for it anymore.

Notice what stopped mattering: whether that work was any good. In the client’s head, the agency’s work had become replaceable. And once something feels replaceable, quality doesn’t get a vote.

A commodity isn’t low quality, it’s undifferentiated

That’s what I call commodity drift.

A commodity isn’t cheap because it’s bad. It’s cheap because the buyer stopped seeing it as different from the alternatives. Gasoline, printer paper, a blog post. The buyer isn’t asking “whose is better,” they’re asking “what’s the going rate.”

Here’s the trap. Sell inputs, the posts and the deliverables and the volume, and you get priced against inputs. That was always true. AI just made inputs feel nearly free, so the drift that used to take years now takes a quarter.

You can see it coming, if you know what to look at

Drift doesn’t announce itself. You don’t lose a pitch. You lose a renewal that was never really in play by the time it came up.

But there are tells, and they show up months earlier.

The client starts asking what individual things cost. A retainer conversation turns into a line-item conversation, which means they’ve started mentally unbundling you. You stop getting invited to the planning conversations and start getting briefed after them. They forward you a competitor’s content and ask “could we do something like this,” which is a request for execution, not a request for a point of view. And the one that should scare you most: they ask for the raw files, the templates, the process doc. Not because they’re leaving. Because they’re starting to imagine doing it without you.

None of those are emergencies on their own. Three of them in a quarter on the same account is the drift, and you still have time to do something about it.

Sell the thing a tool can’t hold

The agencies navigating this well aren’t producing more, faster. They’re selling something harder to copy: real outcomes, and an understanding of the business a tool doesn’t have.

Three places I’d push if you can feel it starting.

Show your thinking, not just your deliverables. The strategic read, here’s what we’re seeing and here’s why we’re changing course, is the one thing a prompt can’t produce, because it requires knowing this specific business. Hand over the reasoning, not just the output.

Tie the work to the outcome they lose sleep over, not the volume you ship. Own their number harder than they do. When you’re accountable to their revenue instead of your deliverable count, “we could just do this ourselves” stops being the obvious move.

Ask the questions AI wouldn’t think to ask. Real curiosity about someone’s business, the follow-up nobody prompted, can’t be automated, because a model only answers what it’s asked.

The contract problem, which is real

The fair objection is that your SOW is scoped in deliverables, your client bought deliverables, and telling them you now sell outcomes mid-retainer sounds like a dodge.

You don’t have to rewrite the contract to fix this. Ship the deliverables you sold. Change what you lead with when you talk about them. If the monthly call opens with the count of what went out, you have trained the client to value the count. If it opens with what you learned and what you’re changing because of it, the deliverables become evidence instead of the product.

Renewal is when you change the paper. The conversation has to change first, or the new paper won’t be believable.

None of this is new, and that’s the uncomfortable part

Everything I just described is what the best agencies have always done.

Skipping it used to be survivable, because good work was enough to carry you on its own. A clean deliverable, shipped on time, kept the retainer alive.

That’s ending. When the inputs are free, being good at the inputs stops being a business. The differentiation you kept meaning to build is now the only thing holding the account, so build it before the client does the math for you.