Agencies stopped hiring the people who learn to catch mistakes. Now their clients are catching them instead.
The logo was wrong. A brand-side marketer at an industry event last month said there was a growing acceptance of “good enough — go,” and then gave the example. The marketer had seen a lot of agency creative in which the brand’s own logo wasn’t right. Before certain campaign work could go live, the marketer had to demand human review.
Human review. Of the logo.
Digiday reported the exchange from a town hall at its AI Marketing Strategies event on September 24. The session ran under the Chatham House Rule, so neither the marketer nor the company was named. The report doesn’t say which tool produced the work, or whether AI touched the logo at all. What it does show is a marketer who could no longer trust an agency to protect one of the most basic elements of the brand, saying so in a room full of people discussing automated workflows and falling standards.
This isn’t the grand science-fiction failure executives were promised. There’s no rogue superintelligence, no machine uprising, and no need to summon a philosopher from Oxford. The logo is simply wrong, and somebody has to notice.
Another participant, from a premium beauty brand, called quality control terrible and said getting brand standards into AI ad platforms was a struggle for all of them. That marketer treats whatever the AI produces as a rough draft and works to a 10/80/10 split: 10% steering at the start, 80% generated by the machine, and a final 10% of the marketer’s own effort to make the result good.
Look closely at that last 10%. It only works if the person doing it knows what good looks like for this particular brand, and that knowledge doesn’t ship with the software.
Several brand-side marketers said their agency partners seemed less motivated to protect brand standards, and that many were leaving platform oversight to junior employees. Hold on to that detail. An agency owner at the same event had something to say about juniors, too.
The owner runs a small-to-mid-sized performance marketing agency. It hasn’t hired college graduates in some time, and the owner doesn’t expect that to change. AI now picks up much of the entry-level work that new hires used to do. The owner was careful to add that this had nothing to do with disliking graduates. It was purely a matter of business.
Here’s the uncomfortable part. Every hire the agency used to make was entry-level, and most of those people grew into some of the best staff it has had.
You couldn’t design a cleaner demonstration of the problem if you tried. The company brought in people with no experience, gave them work, and watched most of them become its best employees. Now automation handles enough of that early work that hiring them looks unnecessary. This quarter’s math may well be right. It just quietly removes the process that produced the agency’s senior people.
A handful of anonymous comments from one conference isn’t a survey, and nothing ties the agency that stopped hiring to the agencies that botched the logo. They don’t need to be connected. The people talking were describing their own businesses, and side by side, those descriptions show an industry cutting the work through which judgment develops while its clients complain that judgment is already missing.
Entry-level tasks are easy to dismiss as production labor. Juniors prepare early drafts, configure campaigns, revise assets, and fix details after someone more experienced explains what went wrong. A lot of it is repetitive, which makes it an obvious target for automation.
The repetition was the education. A junior who fixes the same kind of error a few times starts spotting it before a manager has to. Someone who watches a client reject an almost-correct execution learns that brand standards go further than whether an asset looks plausible. Experience piles up through small decisions whose importance only becomes obvious once they go wrong.
AI can produce the early output. It can’t produce the experience in another person.
If a system completes the task and a senior employee signs off, the agency saves time today. If the junior position disappears, nobody climbs through that task toward becoming the senior who later reviews the system.
One participant at the event offered a line that was meant as encouragement: “Someone who knows AI and has your experience is coming to replace your job.” Fair enough. Where does that someone get the experience?
Another participant saw where this ends. Agencies that stop hiring graduates will run short of expertise once their senior staff retires, and the answer is to rebuild junior roles instead of deleting them. They framed it as a learning-and-development problem more than a technology one. Others said companies have a responsibility to develop young talent through formal mentorship, AI training courses, and internships that pay a livable wage. One exec described pairing older staff, who know how the work actually gets made, with younger staff who are more at ease with the new tools.
It all depends on treating apprenticeship as an investment. Automation pushes the other way, because it makes junior labor look like a cost you can take out of the current quarter.
Some firms are questioning the value of junior employees because machines can do junior tasks. Brand marketers, meanwhile, worry that agencies already leave too much AI oversight to junior employees. The industry wants fewer beginners and better reviewers, apparently hoping the second group will materialize without passing through the first.
Nearly every AI policy I read promises human oversight. Almost none say which human, or how that person learned to tell a good asset from a plausible one.
This story answers the question the policies skip. Increasingly, the human is a junior who was handed the platform, at an agency that has stopped hiring the people who would one day replace its seniors. Sometimes it’s nobody, until the client insists.
Senior judgment takes years of knowing the clients, the audiences, and the unglamorous ways campaigns go wrong. A reviewer has to know when a result is technically acceptable and commercially foolish, or when an asset that passes a platform check still violates the brand.
Cutting early-career work changes the supply of that judgment. The damage is easy to ignore because the missing expertise doesn’t show up the day an automated workflow goes live. It shows up years later, when the experienced people leave, and the organization discovers it optimized away their successors.
A small agency chief, speaking in a separate anonymous session, said AI can shrink campaign implementation from weeks to hours, then asked out loud what value employees add on top of that. It’s a fair question, and any business under pressure to produce more without adding headcount will find the speed hard to resist.
It also makes it easy to confuse finished with good. When a system turns work around in hours, careful review starts to look like the slow part. Generating another asset gets cheaper, but someone still needs to look at every asset.
Output grows. The attention available to examine it doesn’t.
The final review is especially exposed because its value is invisible when it works. Catching a bad logo prevents an embarrassment nobody ever sees. Rejecting generic creative protects brand equity in ways that never show up on a campaign dashboard. A good review looks exactly like a delay, right up until it’s missing.
Premium brands have the most to lose, because their pricing depends on consistency and trust. A beauty company can’t credibly sell meticulous care with ads that mangle its own identity. A faster production process is little comfort once a customer notices.
So “good enough” is a decision. Agencies make it when they use automation to cut labor, raise volume and shorten delivery while quietly reducing the experienced attention each piece of work gets. It holds up fine until enough small failures make the slide impossible to ignore.
Redesigned junior roles could work. New hires can learn by evaluating AI output, working out why it failed, and watching experienced colleagues make the calls the system can’t. That still means hiring people before they have senior judgment, having experienced staff teach them, and accepting a stretch where training costs more than automation alone.
The apprenticeship ladder can be redesigned, but it can’t begin halfway up.
As these comments describe it, the industry wants the benefits of development without paying for development. Agencies want the efficiency of automating junior work, the convenience of handing oversight to junior staff, and the comfort of knowing experienced people will always be around to protect quality. They can’t have all three.
The wrong logo is a small error, which is exactly why it belongs at the center of this story. It’s the kind of mistake a functioning institution catches without drama. Remove the people who learn by catching it, thin out the supervision around automated production, reward speed over scrutiny, and “good enough — go” stops being a temporary lapse in standards. It becomes the succession plan.