AI Can Run Your Marketing Stack, But Someone Still Has To Know What Good Looks Like

The pitch is seductive. Connect an AI tool to the website, the ad accounts, the content calendar, the reporting, then step back and let it run. The catch is that the tool always hands something back, and when no one on the team knows the work well enough to check it, that something can be confidently, invisibly wrong. That is not how the people getting real leverage out of AI work. They are logging more hours than before, and spending the speed on the tasks they already know cold.

Jonathan Giner is the Founder of Marsense and a digital marketing strategist who has spent his career building the campaigns and teams meant to turn marketing into measurable revenue. His work runs across paid media, SEO, lead generation, conversion optimization, and marketing automation, and lately most of his attention has gone to how AI actually changes the day-to-day of that work. What he keeps noticing is that the operators using it well are not coasting.

"'We'll just have AI do it' is probably the biggest lie of 2026," says Giner. He is enthusiastic about what the tools can deliver, and his objection is narrower than it sounds. AI produces real value, and whether a team gets any of it depends on someone who can direct the work and catch what comes back wrong.

What AI still can't see

Giner's core claim is that AI cannot rescue a team from work it doesn't understand. The model always returns an answer, but whether that answer matches reality depends entirely on the person reading it, and without a baseline of expertise a team can't tell a good output from a merely plausible one, the kind of judgment machines still lack. "You cannot manage what you don't understand," says Giner.

The comparison he reaches for comes from outside marketing. A power saw in the hands of someone who has never framed a wall is a faster way to ruin lumber, and the craft is what makes the tool worth owning. "I compared it to a power saw that you'd buy at a big box store," says Giner. "It doesn't necessarily mean that you're a carpenter."

He points to a routine SEO example. Ask an AI tool to check a page and it may report the schema markup is missing, when a look at the code shows it was there all along. Challenge it and the tool backpedals, admitting it couldn't see the markup for one reason or another. "If it hadn't been for your knowledge that the page already had schema, you might have gone in blindly and added a duplicate," says Giner. The blind spot itself isn't the real risk. The risk is not having anyone on the team who can catch it.

What efficiency really means

The speed feels like progress, and the trap sits underneath it, because a team that moves fast can drift away from the judgment that made its work good in the first place. "There's a dopamine hit that comes from leveraging AI for tasks that used to take us minutes or hours, and now we're doing them in a fraction of the time," says Giner.

The word efficiency does a lot of hiding, in Giner's telling. Type a one-line request into a model, paste the result onto the site, and call it done, and by the clock it looks efficient. By any measure that matters it may be the opposite, if the piece uses the wrong keywords, skips the internal links, and never ties back to outcomes. "From a time perspective, it was very efficient," says Giner of that workflow, "but it could be very inefficient," because the output never did the job the page existed to do.

The efficiency he actually means is narrower and more durable. The processes a team has run for years, the ones they can already judge by eye, are exactly the ones AI can now run faster and often a little cleaner. His advice to brands is to start there rather than reaching for work they've never touched. "If you've never done SEO on your own, if you've never managed your own pay-per-click campaign," says Giner, "put that on the back burner," and instead point the models at workflows the team already owns. McKinsey frames the same shift as one marketer overseeing networks of agents rather than a role you quietly delete.

The limits of stacking agents

The cheap part now is standing up agents. One becomes five becomes fifteen, and the arithmetic tempts you into expecting fifteen times the output. Marketing doesn't run like a manufacturing line, though, and a team that treats it that way ships fifteen versions of the same off-brand work. "We want as much quantity as we can possibly get, we just want to make sure it's done right," says Giner. Producing content at that scale only helps if someone is accountable for what comes out the other end.

For Giner, that accountability has to be a person, inside the company or brought in, who reviews what the agents produce and decides whether it's good enough to ship. He's watched operators hand their entire inbox to a fully automated agent with no human checking it, the kind of setup that films well for a video and works badly in practice, and a reminder that more agent autonomy isn't free. "I still see AI as an assistant," says Giner. "It does the work of a team member, or multiple team members." Move from three content creators to one, he suggests, and that one had better be your best.