AI writes the copy, builds the pages, and optimises the bids. So what exactly is your marketing team for? The answer is more important than you think.
Machines write ad copy that converts. They build landing pages overnight. They segment audiences, optimise bids, generate product photography that never required a studio, and draft email flows in the time it takes you to brief a copywriter.
The infrastructure investment behind this is $185 billion and climbing. 160,000 developers are building what the industry now calls "digital employees" — not assistants, not chatbots. Employees. One company shipped a fully autonomous software factory. The code writes itself, tests itself, deploys itself.
All this begs the question, if you run a Shopify brand and AI can handle your ad copy, your email sequences, your product descriptions, and your landing pages and more — what exactly is your marketing team for?
The answer, it turns out, matters more than most people realise. And getting it wrong has consequences that are already showing up.
What's clear for all to see is that the execution layer — the part that used to require a team — is being compressed into a prompt.
Shopify Magic generates product descriptions. Klaviyo builds predictive segments. Your ad platforms auto-optimise creative. AI writes subject lines that outperform your best copywriter. It codes responsive email templates. It produces social content calendars. It analyses customer behaviour across touchpoints and suggests next actions.
Most of what a junior marketer did three years ago, a well-configured AI agent does now. Faster, cheaper, and without taking Fridays off.
The interesting thing is the gap between perception and reality. 71% of enterprises say they use AI agents. Only 11% have them in production. That gap is enormous — but it's closing fast. And when it closes, the brands that treated AI as a novelty will discover they've been lapped by the brands that treated it as infrastructure.
How do deploy people versus machines.
Work is hard for different reasons. And most people lump them together, which is why most AI strategies are incoherent. There are roughly six dimensions that make any piece of marketing work difficult:
AI is exceptional at the first four. Volume, speed, precision, pattern recognition — these are where machines have already overtaken us. They don't get tired. They don't get bored. They don't miss the detail in row 47,000 of the customer export.
The last two — novelty and judgment — are where things get interesting. And in marketing, those happen to be the dimensions that determine whether a brand becomes distinctive or becomes another commodity drowning in AI-generated sameness.
When execution is automated, four human capabilities become the entire game. This framework is borrowed from Nate's Notebook — adapted here for marketing.
Knowing what's good. Not in the abstract — in the specific. Knowing this subject line will land and that one won't. Knowing the difference between a landing page that converts and one that merely looks professional. AI can generate a hundred options. Someone has to pick the right one. In marketing, taste is the difference between a brand with a voice and a brand that sounds like everyone else's ChatGPT output.
Understanding your market — the quirks, the unspoken rules, the things that don't show up in the data. Knowing your customers respond differently in January than March. Knowing a particular segment will hate this campaign even though the numbers say they should love it. This is deep, hard-won pattern recognition that only comes from years inside a specific category.
The ability to learn new tools, new systems, new contexts at speed. When the landscape shifts every quarter, the most valuable skill isn't what you know — it's how fast you learn the next thing. For marketing teams, this means treating every new AI capability not as a threat but as leverage to be understood and deployed. The people who ramp fastest compound their advantage.
The willingness to see what's actually happening rather than what you want to see. To admit the campaign isn't working. That the creative is mediocre. That the strategy needs rethinking. AI is optimistic by default — it tells you what you want to hear. Someone has to be the person in the room who says: this isn't good enough. That person is more valuable than ever.
Building used to be the hard part. Coding the landing page. Designing the email template. Writing the product descriptions. Executing the media plan. All of that is now trivially easy — or getting there fast.
The new bottleneck is specification. Describing precisely what to build.
This is a bigger shift than it sounds. For decades, marketing teams were organised around execution. You hired people who could do things — design, write, code, manage campaigns. The org chart was built for throughput.
But when execution is essentially free, the constraint moves upstream. Now you need people who can define the right thing to build. Who can write the brief that an AI agent can actually execute against. Who can specify the customer journey with enough precision that automated systems can deliver it.
The quality of the specification is the variable. Between a prompt that produces mediocre output and one that produces something genuinely useful — the difference is entirely in how well the human described what they wanted.
And describing what you want, it turns out, requires you to actually know what you want. That's harder than it sounds. It demands clarity about your brand, your customer, your positioning, and your strategy. The woolly brief that a talented designer used to interpret charitably? An AI agent will execute it literally. Garbage in, garbage out — at scale.
80/20 is a useful frame here. In marketing teams getting this right, AI does the execution which is 80% of the work. Humans do the specifications at the start (10%) and fine-tuning at the end (10%). Humans do only 20%, but it's that 20% that determines whether the other 80% has any value at all.
That 20% is: deciding what to do, specifying how to do it, and judging whether it was done well. Strategy. Specification. Quality control. Everything else — the execution, the production, the implementation — is being compressed into agents and automation.
The point in humans is not doing. The point is knowing.
Knowing which campaign to run. Knowing which customer segment is being underserved. Knowing when the data is misleading you. Knowing that the brand voice has drifted. Knowing that this particular creative is right and that one misses by a mile.
There's an inconvenient truth at play here: the window to build these capabilities is short. The brands that figure this out in the next twelve to eighteen months will have a structural advantage — better systems, sharper judgment, faster learning cycles. The ones that don't will still be using AI as a fancy autocomplete while their competitors are running it as an operating system.
The point in humans has never been our ability to execute. It's our ability to know what's worth executing. That was always the point. We're just finally being forced to prove it.
The shift from execution to specification is already underway. The question is whether you lead it or react to it.