Somewhere in a marketing department right now, a perfectly competent brand strategist is debugging a custom GPT at 11 p.m. instead of doing what she's actually good at. She's not doing this because she loves it. She's doing it because her performance review now includes a checkbox for "AI building," and she's terrified of being the person who didn't climb aboard.
We've been here before. A roundtable participant at the AUTONOMOUS conference in September put it perfectly: during the dot-com era, he was told "one day everybody's going to be writing HTML." In the 30 years since, he's written about three lines of it. Instead of building the technology, he used what other people built. The same pattern is playing out with AI, except this time the pressure is louder and the stakes feel higher.
The Performance Review Trap
In , Shopify CEO Tobi Lütke told employees that "reflexive AI usage is now a baseline expectation" and that it would be reflected in performance reviews. Teams now need to prove a job can't be done by AI before asking for more headcount. Plenty of companies followed suit. The bar has moved from using AI to building with it: custom GPTs, agents, automations stitched between the CRM and the email platform.
The result? People are building. A lot. Dataiku surveyed 685 CIOs in July, and 84% agree employees are creating AI agents and applications faster than IT can govern them. Another 72% say they can't reliably verify whether those agents deliver the results they were designed to deliver.
This is Goodhart's law in action: when a measure becomes a target, people start gaming the system, making it less useful as a measure. We're rewarding the act of building, not the outcome of building well.
The Monoculture Problem
The people who lean hardest into building, especially early, tend to share a personality. They're innovative, comfortable with risk, happiest with a blank canvas. Those are wonderful traits. Reward only those traits, though, and you end up with what Melissa Reeve calls a "builder monoculture": a team hired, trained, and promoted around one personality type.
A partner at a small industrial marketing agency told a roundtable his four-person team is all in on AI. How did he get there? "The ones that weren't are not in my agency anymore."
Then he talked about his designer, the slowest adopter on the team. That designer brings something the builders don't: the eye for what's off, the instinct for when the AI output is technically correct but emotionally wrong. Every marketing team needs people who can look at an agent's first draft and say "no" with reasons attached.
Conductors, Not Carpenters
Jessica Vose at Kana uses an analogy I keep coming back to: conductors don't spend their time carving violins from maple wood or restringing the harps. They don't need to be the best oboe player in the room. They do need to know what good sounds like, and they do need to know how to coach their musicians to play their best using the instruments they have.
The future CMO is closer to a world-class conductor than an instrument maker. The job is orchestration, not carpentry.
This doesn't mean marketers can ignore AI. It means the skill that matters is knowing what to ask for, recognizing when the output is wrong, and understanding how the pieces fit together. Square 1 AI's research found that the four skills that actually change a marketer's output are:
- Writing a brief a model can execute
- Building one automation in a no-code tool
- Reading model output with the right suspicion
- Knowing what data must never leave the building
None of that requires you to become a developer.
The Taste Gap
Daniel Kravtsov at Improvado has watched this experiment play out across his enterprise customer base. The pattern is the opposite of the pitch. The best operators of AI agents are not the AI-native juniors. They're the people who spent ten years in spreadsheets, ad accounts, GTM rooms, and agency floors. They look at an agent's output and immediately know which line is wrong. They re-prompt. They re-scope. They override.
The AI-native team without senior judgment ships the same brand-safe, well-formatted, on-trend, statistically average work. Just faster.
The agent doesn't lower the bar for taste. It raises it. And taste, it turns out, is the part you can't onboard in a quarter.
The Builder.ai Lesson
Remember Builder.ai? The company raised over $450 million, including from Microsoft. Their AI assistant "Natasha" was marketed as capable of building 80% of apps automatically. "The majority of the work is done in the first couple of hours," the CEO said.
They filed for bankruptcy. Behind the scenes, 700 human engineers in India did most of the work. The AI hype masked human labor. The promise of "everyone is a developer" crashed into reality: architecture, edge cases, and production hardening still require people who know what they're doing.
The same principle applies to marketing. You can't prompt your way into strategic judgment. You can't automate your way past the need for someone who's seen a few hundred mediocre campaigns and knows what mediocre looks like.
The Real Roles We Need
Marketing teams need more than builders. They need maintainers who keep the tools running. They need monitors who watch for drift, bias, and broken outputs. They need users who put the tools to work in ways that actually serve customers.
Ben Laube puts it bluntly: the real bottleneck isn't a shortage of AI builders. It's a shortage of operators who can run AI systems reliably, integrate them into workflows, and know when to buy instead of burning months building.
The smart sequence? Buy first. Build later. Validate with off-the-shelf, then invest in custom where it actually matters.
What This Means for Your Team
If you're a marketing leader, stop asking "who on my team is building with AI?" Start asking "who on my team knows when the AI is wrong?"
The person who can look at a generated campaign and spot the tone-deaf line, the factual error, the brand inconsistency: that person is worth more than the person who built the automation that produced it.
Marketing is like dating. You don't propose on the first ad impression. And you don't hand the relationship over to a bot just because the bot can send messages faster than you can.
The builders will always get the conference keynotes. But the maintainers, the monitors, the users with taste: they're the ones who keep the whole thing from falling apart.
