Do AI Ad Generators Replace Your Media Buyer? An Honest Agency Answer

Renu owns a six-person performance shop that runs paid media for a stack of Shopify brands. Last week a client forwarded her a question that’s been keeping a lot of agency owners up at night.

“Just got asked: if PMax and AI Max do the bidding, targeting and creative for me, why would I still pay a specialist?” She sent us that on a Slack DM, half venting, half genuinely asking. It’s a fair question. It deserves a real answer, not a defensive one.

Here’s the honest version, the one we give clients and the one we’d want if we were the merchant writing the check.

The automation is good. Better than most agencies admit out loud. And it still doesn’t do the job a good buyer does, just the parts that used to look like the job.

What the automation genuinely does well now

Let’s not insult anyone’s intelligence by pretending the tools are bad. They’re not. They’re the best they’ve ever been, and they’re improving on a quarterly cadence.

Automated bidding is the clearest win. Setting manual CPCs across thousands of keywords or audiences was always a losing game against a system that adjusts bids per auction using signals no human can see in real time. Google’s Performance Max and Meta’s Advantage+ campaigns genuinely beat hand-tuned bidding for most accounts most of the time. A buyer who still sells “I’ll optimize your bids by hand” is selling a buggy whip.

Creative generation has caught up faster than anyone expected. The tools will spin up dozens of headline and image variants, test them, and kill the losers without a human babysitting the rotation. For a brand that was shipping one new ad a month because that’s all the bandwidth allowed, an AI creative pipeline is a real unlock. More shots on goal, automatically.

Audience discovery is the third one. The old craft of manually stacking interest layers and lookalikes has largely been eaten by systems that just find buyers from a conversion signal. Hand a clean signal to Advantage+ and it’ll often find pockets of demand a human would never have thought to target.

So if the pitch is “we push the buttons better than the robot,” that pitch is dead. The robot pushes the buttons fine.

Where the machine still face-plants

The automation is excellent at optimizing toward a goal. It has no idea whether the goal is correct. That’s the whole crack in the floor.

Feed Advantage+ a conversion event set to “add to cart” and it will ruthlessly buy you add-to-carts, including from people who’ll never check out, while your actual profit quietly bleeds. The system did exactly what you asked. You asked for the wrong thing. It has no mechanism to notice, because it can’t see your margins, your repeat-purchase economics, or the fact that this SKU has a 40% return rate.

Creative automation has a sibling problem. It optimizes for the click and the immediate conversion, which means it drifts, relentlessly, toward whatever performs in the next 48 hours. Left alone, that’s how a premium brand ends up with discount-coded, urgency-screaming ads that print short-term ROAS and erode the brand it took years to build. The model has no taste and no memory of who you’re supposed to be.

Then there’s the data going in. PMax and Advantage+ are only as smart as the conversion signal you feed them, and most accounts feed them garbage: deduplicated wrong, counting the same purchase twice, optimizing to a pixel event that fires on the wrong page. Marco, a dev we work with, called us about a “broken” PMax account, we pulled the tag manager, the conversion was firing on page-load instead of purchase. The machine wasn’t broken. It was perfectly optimizing toward a lie.

And none of these tools will ever tell you to spend less. Their incentive is your budget.

The judgment layer no model ships with

Strip away the button-pushing and what’s left is the part that was always the actual job. It just used to be hidden underneath the busywork.

Someone has to decide what success even means for this brand this quarter. Is it new-customer acquisition at a loss because the lifetime value justifies it, or is it blended profitability this month because cash is tight? That’s not an optimization problem. It’s a business decision, and it changes which “goal” you hand the machine. Get it wrong and the most sophisticated automation on earth will efficiently drive you off a cliff.

Measurement is the other half. Knowing whether the ROAS the platform reports is real, or whether you’re paying to retarget people who’d have bought anyway, requires incrementality thinking the platforms have every reason not to do for you. A good buyer runs geo holdouts, reads blended metrics, and catches the moment the dashboard and the bank account stop agreeing.

What’s not on the list of things AI replaces: knowing when to ignore the dashboard. The judgment to say “this campaign looks great and we’re killing it because it’s cannibalizing organic” is exactly the judgment a system optimizing the dashboard will never have.

Repositioning around the tooling, not against it

The agencies panicking about this are the ones whose entire offer was the manual work the automation now does. Of course they’re scared. Their product just got commoditized.

The agencies that are fine, and some are growing because of all this, repositioned. They treat PMax and Advantage+ as power tools, not competitors. The pitch changed from “we’ll run your ads” to “we’ll decide what the machine optimizes toward, feed it clean data, guard the brand, and tell you the truth about what’s actually working.” That’s a harder thing to sell and a much harder thing to replace with a button.

Practically, that means owning the conversion data layer, the thing most in-house teams get wrong and most automation silently depends on. It means setting and revisiting the strategy the automation executes. It means being the brand’s taste, the human who vetoes the discount-screaming variant the algorithm loves. So the work didn’t disappear. It moved upstream, where it was always more valuable anyway.

A shop that makes that shift doesn’t compete with the AI. It operates the AI on the client’s behalf and takes responsibility for the outcome, which is the one thing a tool can’t do.

Proving you beat “just run Advantage+”

Here’s the uncomfortable part, and the part that separates agencies worth paying from agencies coasting. If a merchant can get 90% of your result by clicking “create Advantage+ campaign” themselves, you don’t have a value story. You have a markup.

So the bar moved. The question isn’t “did the account perform,” it’s “did it perform better than the client would have done by just turning the automation on and walking away.” That’s the incremental value, and you have to be able to show it.

We tell agencies to make this concrete. Run the automation-default version as the honest baseline. Then show what your judgment added on top: the margin-aware goal that stopped the add-to-cart bleed, the feed fix that lifted PMax efficiency 30%, the brand guardrail that kept ROAS healthy without torching positioning, the incrementality test that proved a “winning” campaign was just retargeting people who’d already decided. Each of those is a number you can put next to the default and say, this is the part the button didn’t do.

A merchant who sees that comparison stops asking why they’re paying a specialist. They can see the gap between the machine alone and the machine plus a human who knows what they’re doing. The agencies that can’t draw that picture, honestly, probably should lose those accounts.

Pricing when the button does the busywork

If you’re no longer billing for hours spent adjusting bids, the percentage-of-spend model starts to feel weird to clients, and they’re right to feel that way. Charging more as spend goes up made sense when more spend meant more manual work. It doesn’t anymore.

The pricing that holds up ties to the value that’s actually scarce: strategy, data integrity, measurement, and accountability for outcomes. Some shops moved to flat retainers for the judgment layer plus a performance component tied to incremental profit, not platform-reported ROAS. Others price the setup and data work as a project and keep a lean ongoing fee for the strategic oversight. The exact model matters less than the principle, which is that you charge for the thinking, not for keystrokes a robot now handles for free.

Clients accept this when you’ve shown them the incremental-value picture from the section above. Without that picture, any price feels like too much. With it, the retainer reads as cheap insurance against handing a budget to an optimizer that can’t see the business.

What we keep telling clients

When a merchant asks “why pay a specialist if the AI does it all,” the worst response is a defensive one. The honest response is to agree with the premise and then redraw it. Yes, the AI does the bidding, the targeting, and a lot of the creative. No, that was never the hard part of the job. It just looked like the job because it took so many hours.

The hard part is deciding what to optimize for, making sure the data feeding the machine is true, protecting the brand from the algorithm’s worst instincts, and being the one person in the room willing to say the dashboard is lying. None of that ships in a tool. All of it determines whether the tool makes you money or efficiently loses it.

If an agency’s only function is running the automation a client could run themselves, the client should fire them, and we’d tell them to. That’s not a threat to good agencies. It’s a filter. The shops that built their value on judgment are having their best year. The shops that built it on busywork are getting the question Renu got.

Renu didn’t argue with her client. She ran the Advantage+ default for two weeks as a baseline, then showed what changed when she fixed the conversion signal, reset the goal to margin instead of revenue, and pulled the three creative variants that were quietly cheapening the brand. The account did 22% better profit than the automation alone. The client renewed at a higher retainer the same week. The question went away because she answered it with a number instead of a defense.

Questions we get every week

Can I just run Performance Max or Advantage+ myself and skip the agency? You can, and for a small or simple account it might be the right call. The catch is that the automation optimizes toward whatever goal and signal you give it, so if your conversion tracking or your goal is off, it will efficiently spend into the wrong outcome. An agency earns its fee by getting those upstream decisions right, not by pushing buttons.

What do AI ad tools actually still get wrong? They optimize toward the goal you set without any sense of whether the goal serves your business, they drift toward short-term performance in ways that can erode a brand, and they trust whatever conversion data they’re fed even when it’s wrong. They also have no incentive to ever tell you to spend less.

How should an agency price its work now that bidding is automated? Tie pricing to the scarce value, which is strategy, data integrity, measurement, and accountability for real outcomes, rather than to manual hours or a flat percentage of spend. Flat retainers for the judgment layer plus a component tied to incremental profit tend to hold up better than legacy percentage-of-spend models.

What should I ask before firing my media buyer? Ask them to show what they added on top of what the automation would have done on its own. If they can point to a margin-aware goal, a fixed data signal, a brand guardrail, or an incrementality test that changed the outcome, that’s the value. If they can’t, the question answers itself.

If you’re weighing whether AI ad tools have made your media buyer redundant, talk to Monkey Man and we’ll show you the gap between running the automation alone and running it with a hand on the wheel.