AI Won't Make You a Great Creative Strategist - Taste Will. Here's the Agency Moat AI Can't Copy

The line you keep hearing is that AI just levelled the playing field. Any founder, any junior, any tool can now spin up a hundred ads before lunch, so creative is basically solved and the people who used to do it are about to be out of a job.

Here's what actually happens when you try it.

You generate the hundred ads. Maybe twenty are usable. Of those twenty, you've got no real way to tell which three are worth money behind them, so you run all twenty, burn a fortnight of budget, and end up roughly where you'd have been with four good ones. The bottleneck didn't disappear. It just moved from "can we make enough" to "can we tell what's any good", and that second question is much harder than the first.

So no, I don't think AI makes you a great creative strategist. I think it makes the average output worse and the rare good judgment more valuable. Those are two different things, and most people are only talking about the first one.

Let me explain what I mean.

The flood is the whole point

Start with the platform, because the platform is why this is happening at all.

Meta's ad system now rewards advertisers who feed it a lot of distinct creative. Not because volume is virtuous, but because their machine pairs each individual user to the single best ad out of everything you've given it. If you hand it four ads, you've handed it four options to choose from. If you hand it eighty, you've handed it eighty. The more raw material it has, the better it can match, so it quietly favours the accounts that bring more to the table.

That's a sensible design from where Meta sits. But sit where the founder sits and you see the trap. The system is actively pulling everyone towards making more, faster, cheaper. And the easiest way to make more, faster, cheaper is to point AI at the problem and let it run.

So the whole industry is now being nudged, by the algorithm itself, into producing enormous volumes of generated creative. Multiply that across every brand in your category and the feed fills up with the same slightly-uncanny voiceovers, the same stock-ish b-roll, the same hooks that sound like they were written by something that has read a million ads but never wanted anything.

That's the flood. It's not a risk on the horizon. It's the default setting now, and it's getting worse precisely because the platform pays you to add to it.

Taste is the thing that's suddenly scarce

When making things stops being the hard part, choosing things becomes the whole job.

I'll be specific about what I mean by taste, because it gets thrown around like it's a vibe. It isn't. Taste is the ability to look at something the machine just produced and know whether it's any good before you've spent a dollar finding out. It's knowing which of the twenty usable ads has the one hook that'll actually stop a thumb. It's catching that an angle technically tested fine but is quietly cheapening the brand. It's the difference between getting the job done and producing something a real person would stop for.

Here's my take: that judgment is the rarest thing in this whole equation, and it's getting rarer relative to demand. Everyone can now generate. Almost nobody can reliably tell good from mediocre at speed, across a flood, under budget pressure. That gap is the job.

And it cuts the other way too. A genuinely strong strategist who picks up AI gets faster and bolder, because they can test more ideas and trust their own read on the results. A weak one who picks up AI just produces more mediocre work, more confidently. The tool magnifies whatever judgment you walked in with. It doesn't supply the judgment.

To put this into perspective with something I see all the time: a brand will come to us proud of how much creative they're shipping, hundreds of variations a month, and their account is flat. The volume was never the problem. They've got eighty ads and no taste layer deciding which eight deserve to scale and which seventy are noise. More fuel, no driver.

The part no model can copy

Here's the bit I think gets missed in the "agencies are dead" conversation.

An off-the-shelf model knows the whole internet and nothing about your business. It has read every ad ever written and watched exactly zero of your customers actually buy. It can't see which of your angles quietly built repeat purchase versus which one bought a spike of one-time bargain hunters. It doesn't know that the comparison ad that flopped for the homewares brand is the exact one that prints money for supplements. It has no memory of last Q4.

What it can't replicate is pattern recognition across accounts. When you've watched dozens of brands run thousands of ads through the same loop, make the creative, spend behind it, read the result, decide what's next, make the creative, you start to see the shapes. You know which hooks tend to fatigue in three weeks. You know which "winning" angle usually masks a returns problem six weeks later. You know the difference between an ad that's working and an ad that's just over-reporting on a seven-day view.

That feedback loop is the asset. A model can spin up a new ad in seconds. It can't spin up a new product line, open a new market, or tell you that the persona lighting up your account this week is the one you should build your next launch around. Those reads come from sitting inside the loop across many brands over years, and they're exactly the edges that don't show up in any prompt.

So the agency question in the AI era isn't "do they use AI". Everyone uses AI, the same way everyone uses a shared doc. The question is whether there's real judgment and real cross-account pattern recognition sitting on top of the generation. That's the moat. The generation is a commodity. The taste and the pattern memory are not.

What I'd actually ask any agency right now

If I were a founder shopping for help in this environment, I'd stop asking whether a team can produce volume. Of course they can. Everyone can. I'd ask the questions that separate a taste layer from a slop machine.

• Show me what you killed last month, and why. Anyone can show you winners. I want to hear the judgment calls, the ad that tested fine but they pulled, and the reasoning. That's where the taste lives.
• What pattern have you seen across other accounts that changed how you'd run mine? If the answer is generic, they're not pulling on cross-account memory. If it's specific, they are.
• How do you decide which of a hundred AI variations actually gets spend? "We let Meta sort it out" is half an answer. I want to hear a point of view on what makes one of those variations worth real money.
• Where does a human override the machine, and where do they get out of the way? The right answer isn't "always trust AI" or "never trust it". It's knowing which is which, and being able to say so.

You're not testing whether they're modern. You're testing whether there's a person in the middle willing to put their hand up and own the call on what's good. In a feed full of generated sameness, that person is the entire value.

Where to from here

My honest read is that the brands who win the next few years won't be the ones who generated the most. They'll be the ones with the best taste sitting on top of all that generation, deciding what's worth scaling and what's just adding to the noise. The tools are about to get genuinely wild. The scarce skill is still the human one of knowing what's good.

If you've been shipping more creative than ever and the account's gone flat, that's usually the tell. It's not a volume problem. It's a taste-layer problem, and it's hard to spot from inside your own brand because you're too close to the work.

That's the bit a Signal/Noise Audit is built to surface. We go through your creative history with fresh eyes and show you which angles are quietly carrying the account, which are noise, and where the judgment calls are being missed, so you can stop paying to add to the flood and start picking the few things actually worth scaling. If that'd be useful, it's there. No pressure either way.