Michael C. Vincent: Let me set the scene — and the scene is a number.
Michael C. Vincent: An analyst at YCC Macro, Tucker, looked at early-career employment in what researchers call the most AI-exposed quintile — that's the top fifth of occupations ranked by how much AI can actually substitute for what the worker does. Software developers. Customer service reps. Work where AI doesn't just assist — it can cover large portions of the task.
Michael C. Vincent: And in that slice, among workers aged 22 to 24, employment dropped 15.2% between late 2022 and mid-2025.
Michael C. Vincent: Every other quintile held. This one didn't.
Michael C. Vincent: Now, a single analyst's cut of the data — you hold it lightly, right? But then Stanford's Digital Economy Lab, working with ADP Research and their — I mean this is a vast pool of matched employer-employee payroll records — they published a paper. Erik Brynjolfsson led it. 'Canaries in the Coal Mine: Six Facts about the Recent Employment Impacts of AI.' And they found a 15 log-point relative employment decline — call it 16% — for workers ages 22 to 25 in the most AI-exposed occupations.
Michael C. Vincent: Same age group. Same occupational tier. Same direction. Arrived at independently.
Michael C. Vincent: That convergence is the thing. That's what makes this worth a serious hour of your time.
Michael C. Vincent: But here's where I want to slow down, because the data doesn't come with a verdict attached. What Tucker found, what Brynjolfsson documented — the same 15-to-16 percent — gets read two entirely different ways depending on who's doing the reading. Either AI is doing something real to the entry points of career ladders, quietly eliminating the bottom rungs for young workers specifically. Or employers are just raising the bar — hiring fewer, hiring more senior, getting more selective. Not collapse. Adjustment.
Michael C. Vincent: The number is the same in both stories.
Michael C. Vincent: That's the tension. And I'm going to hold it open — because if I hand you a verdict before you've felt the full weight of the evidence, I've done you a disservice.
Michael C. Vincent: Here's where the story gets slippery though — because the macro numbers don't look like a crisis.
Michael C. Vincent: The Yale Budget Lab puts unemployment at 4.2%. The labor market closed Q2 2026 still standing. And Sam Altman — who once told anyone who'd listen that entire job categories would be 'totally, totally gone' — stood up at a conference in Sydney in mid-2026 and said he was 'delighted to be wrong.'
Michael C. Vincent: Delighted.
Michael C. Vincent: So if the apocalypse was misdiagnosed — and it seems it was — why are we still here?
Michael C. Vincent: Because 4.2% unemployment is a national average. And national averages are very good at hiding specific, structural ruptures.
Michael C. Vincent: Consider the scapegoat argument — and it deserves a fair hearing. Eighty-six tech firms cut more than 80,000 workers in a single quarter, Q1 2026. The highest quarterly total in three years. Experts looked at the timing, the scale, the sheer variety of companies involved and said: this looks like post-pandemic correction. This looks like macroeconomic pressure. AI is a convenient story executives retrofit onto forces that were already moving.
Michael C. Vincent: That argument is not wrong.
Michael C. Vincent: But here's what it doesn't explain. The 15-to-16 percent decline isn't in layoffs. It's in hiring. Brynjolfsson's data, ADP Research's matched employer-employee records — what they captured was employers simply not reaching down as far as they used to.
Michael C. Vincent: Fewer offers extended. Fewer first jobs begun.
Michael C. Vincent: The hiring selectivity argument — that AI lets employers do more with fewer, more senior people — that's real. The question is whether new roles eventually offset that compression. Gartner says yes, by 2028, AI starts creating more jobs than it eliminates. And Indeed's Hiring Lab found a 170% growth in generative AI job posting mentions from January 2024 to January 2025. More than 60% of those postings are outside traditional tech companies.
Michael C. Vincent: New roles are forming. That part is real too.
Michael C. Vincent: But those roles — who gets them? Not the 23-year-old who needed an entry-level position to build the résumé that qualifies them for the mid-level position that leads to the senior position. The pipeline assumes rungs.
Michael C. Vincent: And right now, the bottom rung is gone.
Michael C. Vincent: Not gone for everyone. Not gone forever, maybe. But for a specific cohort, in a specific tier of work, in this specific window — you cannot grab a rung that isn't there.
Michael C. Vincent: That's not an apocalypse. But it's not nothing, either.
Michael C. Vincent: The macro looks stable. The macro ALWAYS looks stable — right up until the cohort that couldn't get started doesn't catch up. And then, ten years from now, someone will look at a generation of workers and wonder why the career trajectories are so compressed, why the experience curves look so flat. They'll find the answer here. In this number. In this window.
Michael C. Vincent: The date to hold onto is 2028.
Michael C. Vincent: That's Gartner's marker. Their projection — and it's specific — is that AI starts creating more jobs than it eliminates somewhere in that year. A net inflection. The curve bends back.
Michael C. Vincent: And the Indeed Hiring Lab data gives that projection some ground to stand on. A 170% increase in generative AI job posting mentions — January 2024 to January 2025 — and more than 60% of those postings sitting outside traditional tech companies. Healthcare. Finance. Manufacturing. Logistics. The expansion isn't waiting for Silicon Valley to lead it.
Michael C. Vincent: That's the best evidence for optimism I can give you.
Michael C. Vincent: But I want to turn the thing over, because there's a seam in the 2028 argument — and it's the same seam we've been tracing all along.
Michael C. Vincent: Gartner also noted — in the SAME report — that 40% of companies have already eliminated roles they deemed obsolete. Already. Past tense. The subtraction is happening now; the addition is projected for later.
Michael C. Vincent: And those new AI-titled roles spreading across sectors … who fills them?
Michael C. Vincent: They require something. They require demonstrated experience with AI tools, judgment built on actual workflow exposure, a track record of having done adjacent work. The 2028 expansion thesis is built on the assumption that workers can retrain and step into those openings — that the pipeline is permeable.
Michael C. Vincent: But the whole problem — the Brynjolfsson finding, the Tucker number, the ADP Research matched records — all of it points to the same rupture: the 22-to-25 cohort never got the entry-level exposure in the first place. They couldn't accumulate what the new roles will require, because the positions where that accumulation happens are the ones that went away.
Michael C. Vincent: The ladder assumes you started climbing.
Michael C. Vincent: So the number I'd watch — it isn't total jobs created by 2028. It's whether entry-level hiring in AI-exposed roles actually recovers before this specific cohort ages out of the early-career window entirely. That's the live question. That's the one that doesn't have an answer yet.
Michael C. Vincent: And that's the seam nobody wants to look at directly. Not the total number of jobs. Not the 4.2% unemployment figure, which is real, which the Yale Budget Lab stands behind, which Sam Altman was apparently relieved to point to. The seam is narrower than any of that. It's a pipeline question. A sequence question. If entry-level hiring in AI-exposed roles stays 15 to 16 percent below baseline — not forever, not catastrophically, just persistently, just long enough — then the workers who would've filled those roles never accumulate what the next rung requires. They don't become the mid-level hire. They don't become the senior judgment that Gartner's 2028 expansion thesis assumes is ready and waiting.
Michael C. Vincent: Expertise doesn't arrive whole. It compounds. You do the entry-level work, you make the entry-level mistakes, and somewhere in that friction is the thing that makes you hireable for what comes next. Brynjolfsson's data — the ADP Research matched records, the controlled comparison — what it captured isn't just a hiring dip. It's that compression happening quietly, at the base. The 22-to-25-year-old who couldn't get the first position can't demonstrate what they'd have learned in it. And the new AI-titled roles spreading across healthcare, finance, logistics — those postings aren't asking for willingness. They're asking for track record.
Michael C. Vincent: So the real question — the one this debate keeps circling without quite saying aloud — isn't how many jobs AI eliminates. It's whether inexperience can still become expertise when the door where that conversion happens is closed. That's the live question. The one the data doesn't answer yet. And the cohort it matters most to isn't waiting for 2028.
