Your Skills Got Sharper. The Place to Practice Them Didn't Survive.

A couple of days ago I came across a tracking report from Stanford's Digital Economy Lab, updated through April 2026 — almost four years of data now. One number stuck with me: for 22-to-25-year-olds working in "AI-exposed" occupations, employment sits 19% below their peers in "low-exposure" occupations of the same age.

The second half of that sentence is the part that actually stings: the gap isn't a one-time drop. It's been widening by roughly half a percentage point every month, for close to four years, with no sign of reverting to the mean.

In plain terms: this didn't happen on a single day. It's been chewed away, month after month, for four years — and it's still being chewed away right now.

I went and checked the original report. The lead author is Erik Brynjolfsson, and the dataset comes from ADP — one of the largest payroll processors in the US — built on actual payroll records, not surveys. They deliberately stripped out the entire tech industry and re-ran the numbers; the effect held. Which means this isn't just "the coding industry eating itself" — there's a more fundamental mechanism underneath.

The mechanism itself isn't complicated. What AI is best at right now is work with clear boundaries, high repetition, and low cost of error — scaffolding code, fixing a simple bug, turning a document into structured data. Who used to do that kind of work? People just starting out. It was the rung you climbed onto to practice. A senior engineer with one junior used to produce two units of output; now a senior engineer paired with an AI can produce more than that, and the junior's spot goes empty. The junior didn't get worse at their job. The rung got pulled out from under them.

I happen to be one of the people doing this rung-pulling — just on my own small scale.

Over the past couple of years, in my spare time, I've built myself a few automation systems: one that files everything I read into a personal knowledge base and tags it; another that picks material, drafts an article, and pushes it to my WeChat drafts folder every day. Late one night last week I was tuning another script, one meant to judge on its own whether a given piece of information had already been logged before. I used to do that comparison by hand, scrolling through old records one by one. Now I hand it to a script, and it sorts through dozens of items in a few minutes — more reliably than I ever managed by hand.

Ten years ago, on an actual team, that job would almost certainly have gone to a new intern to cut their teeth on: scan the records, compare them, flag the duplicates. Tedious, low barrier to entry, exactly the kind of thing you'd use to get someone familiar with how things work. Now there's no one left to hand it to. I wrote a script, and in one evening, I dismantled that particular practice rung myself. Honestly, it wasn't that I set out to tear it down — it's just that the work genuinely wasn't worth anyone's time anymore.

Setting aside the "wow, AI is impressive" reaction, there's something else I actually want to say clearly: what I did inside my own little system and the macro trend in Stanford's data are the same mechanism, just at wildly different scales. The rung I pulled out by writing a script in my spare time is, scaled up across an entire industry, that 19% and that half a percentage point a month.

My generation grew up hearing one line a lot: get good enough at a craft and you'll never go hungry. That line has a hidden premise — you need somewhere to actually get good, some rung to practice on. As long as that rung exists, the line holds. Once it's pulled out, the question changes entirely: it's no longer "are my skills sharp enough," it's "do I have a path forward that doesn't depend on that rung, or on any single employer."

These two questions sound similar, but they're actually completely different problems. The answer to the first one is: learn another language, grind a few more practice problems, touch up the resume again. The answer to the second one looks more like: do I have something in hand that proves my value without a company signing my paycheck — a tool I maintain myself, a product that serves even just me, a stream of income I can settle independently of any employer. The handful of automation systems I built for myself never started as some grand plan. I just wanted to save myself some time. Looking back, they happened to build me that second path along the way.

The data is still trending down, and I have no idea when it stops. The only thing I can confirm right now is this: that rung really has been pulled out, it's still being pulled, and it's not going to grow back easily.