Amazon's algorithm already knows your customer better than you do. The question is whether your listings are built to work with that, or against it.

Vincenzo Toscano from Ecomcy is hosting a free live webinar with four speakers tackling this from every angle: how Alexa sees your listing, how to win AI assistant traffic, why your launch targeting is off, and more.

📅 Tue, Sep 1st | 12 pm EST / 6 pm CET

Register for the free webinar here

The debate about whether AI is destroying entry-level jobs has been sucking up all the oxygen for months now. It's also the wrong debate. Not because the answer doesn't matter, but because even if you settled it conclusively — yes, no, it's complicated, here's a spreadsheet — you'd still be staring at the bigger problem nobody wants to talk about.

Entry-level work was never just about the work. It was the mechanism through which people developed judgment. And AI is hollowing out that mechanism whether the job title survives or not, at the exact moment every ecommerce operation on the planet is becoming more dependent on humans who can tell when a machine is confidently, expensively wrong.

For anyone running a team that touches product listings, ad spend, or inventory, this isn't a philosophical concern for a panel discussion. It's a staffing and risk question landing in your next hiring round.

The data is genuinely alarming in places, so let's be honest about it rather than pretending otherwise.

An analysis of payroll records covering millions of American workers found that people aged twenty-two to twenty-five in AI-exposed occupations saw a roughly sixteen percent relative decline in employment after generative tools went mainstream — while older workers in identical roles held steady. The Federal Reserve Bank of New York reports that unemployment among recent graduates now sits above the rate for the workforce as a whole, with underemployment near forty-two percent. A separate study of nearly eight thousand salaried postings found the share open to candidates with zero experience fell by around seventy-three percent over four years. To roughly one in fifty. And not just in tech — healthcare, government, utilities and transit all showed the same contraction.

Set against that: some labour economists point out that the weakening market is hitting graduates and non-graduates alike, which makes it hard to pin the damage cleanly on AI versus a broader hiring slowdown. One study of more than twenty thousand firms found that the heaviest AI adopters actually grew their entry-level headcount after adoption.

The honest read is that the job-count effect is real, early, and unevenly distributed. Reasonable people are still arguing about how much is AI and how much is the economy being the economy.

But here's the thing. Whichever way the counting argument resolves, it's a surface reading. Fixating on how many junior roles survive completely misses what those roles were for.

Think about what a junior person on an ecommerce team used to do. They pulled keyword lists by hand. They reconciled settlement reports line by line. They triaged the customer service queue. They wrote the first, clumsy draft of listing copy that someone senior then tore apart with the enthusiasm of a disappointed parent reading a school essay.

Almost none of it was glamorous, and a great deal of it is now automatable. That's precisely the problem — because the value of that work was never the output. It was the friction. In the act of doing the boring thing badly and then, over months, doing it well, a person builds pattern recognition. They develop what experienced operators sometimes call "the smell" (a technical term I just made up, but you know exactly what I mean).

It helps to separate two capabilities that look identical from the outside but are nothing alike underneath. The first is fluency in operating AI: knowing how to prompt it, chain outputs together, move quickly. The second is the deep, internalised understanding of a domain that lets you glance at a plausible AI output and feel, before you can articulate why, that something's off.

The first capability is genuinely easy to acquire — most people pick it up in an afternoon. The second forms slowly, through years of doing the work yourself, including the parts now being handed to the machine. When AI performs the grunt work, throughput looks wonderful. The second capability quietly fails to develop. And nobody notices, because the dashboards tracking volume rather than validity keep looking healthy (right up until the moment they very much don't).

The unsettling part is that the second capability is exactly what supervising AI requires. You cannot catch a mistake you lack the knowledge to recognise. So the more heavily an organisation leans on AI, the more it needs people with deep independent judgment — at the very moment it's automating away the roles that used to manufacture them.

Businesses are coming to depend on a resource they're simultaneously ceasing to produce. Nobody's deciding this on purpose. Each individual choice to let the tool handle the junior work is rational. The cost only shows up later, and it shows up on someone else's balance sheet.

Compounding all of this is what's happening upstream, in education, before anyone reaches your team at all.

A study of more than six hundred participants in the UK found a clear negative relationship between frequent AI use during study and critical-thinking scores, with the effect running through what researchers call "cognitive offloading" — the habit of delegating mental effort to an external system. The youngest users were the most dependent, and higher education offered only a partial buffer. A separate experiment measuring brain activity while people wrote essays found that those using a generative tool showed the lowest levels of neural engagement of any group (which is a polite way of saying the tool was doing the thinking and the brain was doing the watching).

The point is not that young people are lazy. Reasoning is closer to a muscle than a fact. It's built by effortful struggle, and if the struggle is outsourced during the years it's meant to form, it doesn't form. What this means in practice is that businesses are increasingly receiving people who are fluent with AI and comparatively thin on independent reasoning, and then placing them into roles where AI does most of the remaining thinking too. The gap doesn't get a chance to close. It gets institutionalised.

None of this would matter much if the stakes were low. In ecommerce, they're not — because the modern operation runs on precisely the kind of plausible, occasionally-catastrophic automation that demands human judgment to police it.

Consider how much of a contemporary Amazon or D2C business is now machine-generated or machine-managed: listing copy and enhanced brand content, ad bid and budget management, demand forecasting, dynamic pricing, customer service responses, creative production, and (increasingly) the content written specifically to be surfaced by answer engines. Every one of these systems produces output that is fluent and, every so often, expensively wrong.

A hallucinated compliance claim buried in a product description. A forecast that's confident and off by a wide margin, leaving you overstocked or out of stock. An advertising budget quietly reallocating spend towards the wrong terms while ROAS looks fine on the surface. A repricing rule that races your hero product to the bottom (always fun to discover on a Monday morning).

Catching any of these requires someone who knows the category, the margins, and the platform's rules well enough to sense that something is off before it becomes a chargeback, a suspension, or a write-off. In a large organisation, that someone is a senior operator. In a small ecommerce team, it was very often the same junior who used to do the manual work and, in doing it, learned to smell trouble.

Automate the manual work away completely and you're left with output and nobody equipped to vet it. Which is fine. Until it isn't.

The deskilling evidence extends well beyond ecommerce. A clinical study found that specialists' ability to detect abnormalities without AI assistance measurably declined after only a few months of working alongside the system. Workplace surveys keep finding that a large share of employees don't routinely check AI output, and that the flawed material slipping through costs real hours to clean up after the fact. This is simply what happens to any judgment that stops being exercised. It atrophies — quietly, invisibly, and on no one's KPI tracker.

The reflex answer is "teach AI literacy," and that answer isn't wrong so much as insufficient. On its own it can make things worse, because most AI-literacy training teaches the easy capability (operation) and skips the hard one (judgment). Turning out a generation that's even more fluent at operating tools it cannot evaluate is not a solution. It's an acceleration.

The more useful reframing — for how you design an entry-level role rather than how you pontificate about education — is to preserve productive friction on purpose. In practice, this is very ordinary: have a junior attempt the analysis, the copy, or the forecast themselves before the AI assists, so the struggle that builds judgment still takes place. Require them to interrogate and, where warranted, overrule the AI's output rather than simply pass it along. Keep some work deliberately unassisted (in the way a pilot still practises manual landings long after autopilot could handle them). Build early-career roles around validation and exception-handling rather than raw throughput, so the job itself trains a person to catch what the machine misses.

The translation for ecommerce operators is blunt: do not let a new hire's entire job be running the tool. Make part of it checking the tool — and being able to explain, in category terms, why it's wrong.

The workforce being shaped by all of this won't look damaged on paper. It'll look faster and more productive than any generation before it, because it will be. The risk is precisely that it looks fine: high output, plausible work, fluent operation of every system you put in front of it — right up until something non-routine breaks and there's no one in the building with the internalised judgment to notice in time.

The question isn't "can this person use AI?" Most of them can. The question is where the person who can catch AI's mistakes is going to come from — and whether you're still building that person, or merely hoping to hire them from a pool that is quietly, and perhaps permanently, getting smaller.

P.S. If you're currently onboarding a junior whose job description is essentially "run the tools and report back," maybe throw in a weekly exercise where they have to do one task the old-fashioned way. Think of it as intellectual cross-training. Their future self (and your future P&L) will thank you.

P.P.S. Yes, the irony of writing this newsletter with AI assistance is not lost on me. In my defence, I did the thinking first. Mostly.

You can help us!

Spread the word to your colleagues or friends who you think would benefit from our weekly insights 🙂 Simply forward this issue.

In addition, we are open to sponsorships. We have more than 66,000 subscribers with 75% of our readers based in the US. To get our rate card and more info, email us at [email protected]

Jo Lambadjieva is an entrepreneur and AI expert in the e-commerce industry. She is the founder and CEO of Amazing Wave, an agency specializing in AI-driven solutions for e-commerce businesses. With over 13 years of experience in digital marketing, agency work, and e-commerce, Joanna has established herself as a thought leader in integrating AI technologies for business growth.

For Team and Agency AI training book an intro call here.