DEV Community

Cover image for The Last Generation That Learned the Hard Way
Debashish Ghosal
Debashish Ghosal

Posted on AI-assisted

The Last Generation That Learned the Hard Way

#ai #employeeexperience #career #healthydebate

In August 2026, Stanford's Digital Economy Lab updated its "Canaries in the Coal Mine" paper using ADP payroll data. Two findings sit side by side:

  1. Employment for workers aged 22 to 25 in highly AI-exposed occupations, software development among them, is about 19% below where it would be if it had kept pace with less-exposed peers. In the July 2025 data it was 15%.
  2. Experienced workers show no comparable gap.

If you're 5 to 10 years into your career, the second point looks like good news.

Read it again, though. Those 22-to-25-year-olds are the people who were supposed to become the next you.

This is the last part of a series on how engineers who learned to build before AI see the AI era. We've covered their trust, the gap with the C-suite, what mandates actually implement, and how company type changes the job. This one is about the future, and it's more question than answer.

The Ladder Is Missing Rungs Below You

The Stanford researchers are careful, so I'll be careful too. They call these "descriptive patterns, not causal estimates." The gaps get smaller when you account for education, and some of the divergence appears before generative AI was in wide use.

The mechanism they describe still matters. The adjustment seems to come mostly from hiring fewer young workers, not from laying people off. Nobody's getting fired. The door is just opening less often.

The counterargument deserves a fair hearing. In January 2026, Oxford Economics said that "firms don't appear to be replacing workers with AI on a significant scale." In Challenger, Gray & Christmas data, AI was cited for nearly 55,000 US job cuts in the first 11 months of 2025, which is just 4.5% of reported job losses. Oxford suspected that some firms are "trying to dress up layoffs as a good news story," covering for things like past over-hiring. They also argued that rising graduate unemployment looks cyclical, partly driven by a glut of degree holders.

Both things can be true. AI may not be causing mass layoffs, and it can still be a convenient reason not to hire the junior developer this year.

When your team last had budget for a new hire, did it go to a junior?

Your Tacit Knowledge Is the Moat. For Now.

The most interesting part of the Stanford revision is a distinction between two kinds of knowledge.

Employment fell among young workers in occupations that rely on codified knowledge: formal, documented material you can learn from textbooks and written procedures. Employment rose among experienced workers in occupations that rely on tacit knowledge: things picked up through practice, mentorship, and repeated exposure to real situations.

Their explanation: generative AI is good at reproducing knowledge that's already written down. Experience-based knowledge is harder to copy.

For a mid-career engineer, the tacit part is everything no doc tells you:

  • Which service falls over when the batch job runs late.
  • Why the retry logic is written that strange way.
  • The instinct that says "this PR is fine but I don't like it" before you can explain why.

You got that from 2 a.m. incidents, reading other people's code, and seniors who were patient with you. That's your moat.

Here's what worries me, and this is my theory rather than Stanford's: tacit knowledge comes from doing the work badly for a while. If junior hiring falls, and the juniors who do get hired let AI handle the work they would have struggled through, where does the next generation's tacit knowledge come from?

There's a small signal in the data already. In Stack Overflow's 2025 survey, 20% of developers said they've become less confident in their own problem-solving because of AI. And METR's 2026 update describes experienced developers who wouldn't take part in a study because they didn't want to work without AI.

None of that is a disaster. But it hints at a possibility worth taking seriously: your generation may be the last one that learned entirely the slow way.

The Promotion Rules Just Changed

The WRITER 2026 survey of 1,200 executives shows what leadership now rewards. These are executives' self-reports about knowledge workers in general, not just engineers, so treat them as intent rather than proof:

  • 77% said employees who don't become AI-proficient won't be considered for promotions or leadership roles.
  • 60% said they plan to lay off employees who can't or won't use AI.
  • Workplace Intelligence found AI super-users were about 3x more likely to have received both a promotion and a raise in the past year.

Put that next to part 1 of this series, where mid-career engineers were shown to be the most careful about trusting AI output. If leadership reads careful as slow, the most calibrated people in the building could get passed over.

Is your organization rewarding AI usage, or AI judgment? Would it know the difference?

The Identity Question Nobody Budgeted For

LeadDev spoke with a staff engineer at a FAANG company who had built an internal framework for measuring AI adoption, loosely based on Steve Yegge's "executive chef" model, where the engineer orchestrates agents instead of writing every line.

The biggest obstacle they found wasn't skill. It was this:

"There is a genuine pushback of 'I don't want my job to be an orchestrator of agents,' which is an identity concern, not a capability one, and a much harder thing to measure or address."

That's worth sitting with. A lot of engineers who started in 2015-2020 chose the career because they liked building things with their own hands. Being told the job is now to direct a kitchen of AI cooks isn't only a skills change. It's a change in who you are at work.

No survey I found asks about that directly. Maybe it should.

Three Bets a Mid-Career Engineer Can Make

This part is opinion. I'd call these reasonable bets, not proven strategies.

Bet 1: Be the verifier. When Stack Overflow asked when developers would still turn to a human, 75% said "when I don't trust AI's answers." Someone has to be the person others trust when the AI is unreliable. That's a real role, and tacit knowledge is what qualifies you for it.

Bet 2: Be the teacher. If junior hiring is shrinking, the juniors who do get hired are worth more, and so is whoever gives them the scar tissue. Pair with them on the incident. Make them debug without the assistant sometimes. The Stanford data suggests that apprenticeship is exactly what AI can't replace.

Bet 3: Be the measurer. In Harness's 2026 survey, 49% of engineers wanted a say in defining how AI productivity is measured. Leadership is going to measure something. Mid-career engineers are the people who can explain why "tokens used" is the new "lines of code," and offer something better.

None of these is "refuse AI." All of them are about making sure the organization values what you already know.

The Takeaway

AI didn't make experience worthless. It made unshared experience worthless.

Last questions of the series, and I'd really like to hear your answers:

  • Who taught you the thing no doc could? Will that person have a successor?
  • If you're 5 to 10 years in: are you mentoring anyone right now, or has AI quietly taken that role?
  • If you're early in your career: where are you getting your 2 a.m. scars?

Thanks for reading the whole series. If one argument in it seemed wrong to you, say so in the comments. That's how this is supposed to work.

Sources

Top comments (0)