Is AI really taking entry-level jobs? We're more worried about the easy work disappearing

Hello fellow nerds,

An old clip of Anthropic CEO Dario Amodei has been making the rounds again, the one where he warns that AI could wipe out half of all entry-level white-collar jobs. So, is AI really taking entry-level jobs? From the numbers we can see, most of those jobs still exist. What seems to be shrinking is the easy work that used to turn a beginner into someone worth promoting, and that worries us more than the clip does.

Amodei first made that warning in a May 2025 interview with Axios, where he gave it a window of one to five years and said AI companies and governments needed to stop "sugar-coating" what was coming. We're now about a year and a half into that window. If you only read last week's headline, you'd be tempted to wave the whole thing away. The October jobs report showed the US adding 165,000 jobs, far more than forecasters expected, and economists lined up to call it the best report in a while.

We felt that relief too, for about a paragraph.

The numbers under the good news

The same report had a quieter line further down. People who lose a job are staying out of work longer. The average stretch of unemployment rose to 27.1 weeks, the longest since September 2021. A strong month for hiring and a long, slow wait for the people still looking can both be true at once, and right now they are.

Then there's the group the clip is really about. The New York Fed tracks recent college graduates, people aged 22 to 27, every quarter. Its latest update has some good news: their unemployment rate fell to 5.2%, the lowest in more than a year. But their underemployment rate, the share working in jobs that typically don't need a degree, ticked up to 42.5%. That's more than two in five young grads with a diploma on the wall and a job description that never asked for it.

We want to be fair about what those numbers can and can't tell us. Nothing in them says AI did this. Hiring slows and speeds up for plenty of reasons, and none of these reports names one. We dug into why the jobs report looks calm last month. The New York Fed itself points out that many of these "non-college" jobs are fairly skilled and well paid, and that a lot of graduates move into better roles after they gain some work experience.

That last part is the bit we keep coming back to. After they gain some work experience. The whole system assumes there's somewhere to get it.

What one AI company found in its own building

The most honest data point we've seen on this came from an AI company describing itself. In October, Anthropic published a report on how its own workforce has changed, and TechCrunch walked through the findings. Since early 2025, the share of its technical hires who are junior fell from 27% to 9%. The company says its own trends could be an early sign of "how AI will shift work organization, hiring, and labor demand across the economy," because it has built its workflows around AI more thoroughly than almost anyone.

One detail stuck with us. In an internal survey, some employees said their job had shifted "from actually doing work to managing Claudes." The company that makes the AI now has people whose main task is supervising the AI, which may be the most 2026 sentence we've ever typed. The report also says "human attention and judgement" is turning into the bottleneck, because the AI produces more text than people can read and sort.

Put those two findings next to each other and you can see the problem. The skill that matters more every month is judgment: knowing when the draft is wrong, when the code will break, when the customer email will land badly. And judgment has always come from doing a lot of small, boring work and getting some of it wrong while someone more experienced was watching.

The job that was also a classroom

We think the entry-level job was always two things in one. It was work a company needed done, and it was a classroom the company didn't have to call a classroom. The junior paralegal reviewed documents. The junior developer fixed small bugs. The new marketing hire wrote the first draft of every product description. Nobody loved that work, but it's how people learned what "good" looks like.

Those are exactly the tasks today's AI tools are best at. So a company can keep hiring juniors and still quietly remove most of what used to teach them. Headcount stays the same, the jobs report stays calm, and the learning dries up where no statistic can see it. The senior people companies will badly want in ten years are supposed to come from today's beginners, and we haven't heard a convincing plan for where they'll get their practice.

We're skeptical of the scariest version of the clip. Grad unemployment just went down, and the economy keeps adding jobs. "Half of all entry-level jobs" in a few years still looks like a forecast to us, and the data don't show it happening yet. But we're uneasy about the slower version, the one where the jobs stay and the training inside them disappears. Nobody sends out a press release when that happens.

What we'd actually do about it

If you're early in your career, use AI for the easy work, and then do some of it yourself anyway, on purpose. Check what the tool gives you as if your name is on it, because it is. Keep notes on the calls you made and why, since that's the proof of judgment an interviewer can't get from your AI-polished résumé. And when you interview, ask plainly how new people learn the job there. The answer tells you a lot.

If you run a small business, this may be a good moment to hire someone junior. Bigger companies are hiring fewer of them, so good people may be easier to find. Give them AI tools and real problems, but make checking and fixing the AI's work part of the job, with you looking over their shoulder for the first few months. You get help now and a trained person later.

If you work alone, you're your own junior. When AI writes every first draft, you skip the reps that made you good in the first place. We'd keep a few tasks you still do the slow way, just to stay sharp.

We don't know how this plays out, and we're suspicious of anyone who says they do. We're hopeful that companies figure out the training part before it becomes a gap, and worried that most won't notice until they need people they never trained.

So here's the question we'll leave you with: if the easy work that taught you your job had gone to AI on your first day, how would you have learned to do it well?

— Artificially Nerdy