Notes · April 22, 2026
The biggest risk of AI isn't that it replaces you, but that it makes even incompetent people feel smart
7 min read
The biggest risk of AI isn't that it replaces you.
It's that it makes even incompetent people feel smart.
Because today all it takes is opening ChatGPT, Claude or Gemini, getting a well-written answer with a confident tone and a clean format, and suddenly too many people feel competent. Not because they actually understood something, but because they're holding an answer that sounds good.
And that's exactly the problem.
AI isn't creating experts, it's giving an (apparently) authoritative voice to people who often aren't experts at all. It's letting shallow people look prepared, weak reasoning look smart, well-written mistakes pass for insights.
And the most dangerous part is that most of them don't even notice. Because when an answer is convincing, people stop checking. And so they don't really use AI to think better, they use it to skip the hardest step of all: thinking.
AI doesn't make you good, it just makes you faster at showing your real level.
If you have method, vision and real skills, it becomes a huge lever. But if you don't have them, it's just an elegant disguise for your mediocrity.
And the market today is filling up with exactly that. Not with real skills, but with false appearances that are just well packaged.
The symptoms in marketing: what bad advice sounds like when it's well written
In marketing and advertising the phenomenon is particularly visible because the people receiving the advice are often as unskilled as the people giving it.
The most common signal: generic advice presented as proprietary insight. "You should test different creatives to find the one that performs best." True, obvious, useless. But written confidently, maybe with a bullet point and some bold text, it passes for strategy.
The second signal: recommendations that completely ignore context. An AI-amplified consultant tells you to increase the budget on the campaigns that perform. They don't know you have an attribution problem, that the pixel is dirty, that you're comparing different conversion windows. They don't know because they didn't ask the right questions. They just gave a right answer to a generic question.
The third signal, the sneakiest: the right advice at the wrong time. AI has no sense of timing. It tells you to run aggressive retargeting on an audience you've already burned three times. It advises you to broaden interests when the real problem is the offer. It's right in theory. In practice, it's useless.
A real professional isn't the one who has the answer. It's the one who asks the questions that take apart the wrong question.
Why clients can no longer judge quality
The problem isn't only on the supply side. It's on the demand side too.
The clients buying advertising services today are increasingly bombarded with content that sounds technical and precise. They've read threads on Twitter, watched videos on YouTube, received newsletters with "frameworks" and "models" that look structured. They feel more informed than before. And in part they are. But they've acquired the vocabulary without the judgment.
They know what a CPM is. They don't know how to judge whether the CPM they're paying is high or low given their industry, their audience, their market stage. They know what an A/B test is. They don't know whether the test structure you're proposing is statistically valid or will produce garbage data.
This semi-knowledge is dangerous. It creates clients who believe they can judge when in reality they're just recognizing the vocabulary. And anyone producing well-written AI output knows that vocabulary perfectly.
The result is a quiet race to the bottom. Clients pick whoever communicates better, not whoever works better. And that pushes the market toward the people who are good at packaging, not the ones who are good at solving real problems.
How a senior Meta Ads professional uses AI, and how a beginner uses it
The difference isn't in the tools. It's in the method that comes before and after using the tools.
A beginner opens AI with an open-ended question: "How can I improve my Meta Ads campaigns?" They get a structured answer, copy it, apply it. They don't have the context to know what's relevant and what isn't. AI has filled the void of method with output that looks like method.
A senior does the opposite. First they analyze the data. First they formulate a hypothesis. Then they use AI to speed up a specific operation: "Give me 10 headline variants for a copy angle based on time savings for an audience of freelancers aged 30-45 with this brief." Not to think. To produce faster what they've already decided to test.
AI becomes a lever when you're the one driving it. It becomes a crutch when you let it drive you.
I've seen this mechanism work very concretely in the market validation work that precedes any serious campaign — as in the SaaS/AI tool case where we used AI to reach 1,240 waitlist signups before writing a single line of code. That wasn't AI replacing reasoning. It was AI in the service of a hypothesis that was already formed.
The selection mechanism that will separate the real from the amplified
In the short term, the market rewards whoever sounds good. In the medium term, it rewards whoever produces results.
The problem is how long the short term lasts. In professional services, the feedback loop is slow. A client hires a consultant, waits 3 months, sees unsatisfying results, concludes that "the market is tough" and switches consultants. They don't always attribute the failure to the wrong consultant. Sometimes they do. But often they don't.
Still, there's a selection mechanism that works anyway, just more slowly. The clients who survive and grow are the ones who worked with real professionals. They become better clients, with better-calibrated expectations, with more ability to spot real value. They start asking the right questions. And the right questions expose the AI-amplified in a matter of minutes.
The market cleans itself up. Just slower than we'd like. And in the meantime anyone doing serious work also has to get better at communicating that they're serious — not to look competent, but to make visible the competence they already have.
How to defend yourself as a client: the questions that tell the expert from the AI-amplified
If you're evaluating a consultant or an agency, there are questions no AI answers well if the person receiving them has no real experience.
"Tell me about a time you stopped doing something that seemed to be working." Real professionals have stories of pivots, of wrong hypotheses, of things abandoned. The AI-amplified always talk to you about successes.
"How do you handle an account that looks fine on Meta's data but bad on actual sales?" This question requires experience with the attribution problem, with conversion windows, with hybrid data models. It's not an answer you improvise by reading generated output.
"What would you never do on an account like mine, and why?" Strong, specific opinions about limits and mistakes to avoid are the most reliable signal of real competence. AI tends toward completeness and balance. A senior has sharp positions.
"What data would you ask me for before starting work?" If the answer is vague or generic, you have your answer. A professional knows exactly what they want to see: campaign history, CRM data, average deal size, lifetime value (LTV), attribution data. They know why they want it and what they'll do with it.
AI-amplified incompetence holds up under general questions. It collapses under specific questions that require embodied experience, not synthesized text.
If you want a strategy session where we talk about your specific case with no pre-packaged answers, you can book it here. It's not AI answering. It's me.
Read also: why ROAS alone tells you nothing about the health of your ecommerce
Davide Cosmai
Meta Ads Expert & Growth Strategist · Meta Business Partner. 15+ years running Meta campaigns. €52M+ in revenue generated for clients.