Comeau argues the junior job collapse isn't a normal downturn — it's a permanent reallocation where the tasks that justified hiring juniors (small tickets, boilerplate, proof-of-concept work) are now absorbed by senior devs using Copilot, Cursor, and Claude Code. He backs this with Indeed's software postings sitting ~35% below the Feb 2022 peak while other sectors have recovered, and LinkedIn data showing junior/entry-level postings down ~50% versus only ~15% for staff/principal roles.
Comeau explicitly breaks from his own two-year pattern of avoiding doomerism to call out the prevailing story that juniors paired with Copilot will match mid-level productivity. As an educator selling courses to juniors, he has commercial incentive to stay optimistic, and his willingness to abandon that diplomacy signals he believes the floor-raising narrative is actively misleading newcomers about their prospects.
Comeau leans on Wozniak's circulating quote and stress-tests it against hard data: Hacker News' 'Who is Hiring' thread shows junior-tagged roles at their lowest absolute count since 2011, and entry-level postings have collapsed disproportionately versus senior roles. The asymmetry between junior and senior posting declines (50% vs 15%) is presented as proof that the barrier to entry, not the field itself, is what's broken.
Josh Comeau — the React educator behind CSS for JavaScript Developers and the Joy of React — published the fifth installment of his WHAM newsletter under the title The AI Elephant in the Room. The piece, which hit #1 on Hacker News with 296 points and a sprawling comment thread, is unusual for Comeau: he sells courses to junior and intermediate devs, and he's spent two years carefully avoiding doomerism. This essay drops the diplomacy.
The framing leans on a Steve Wozniak line that's been circulating in dev circles for months: "The hardest job in tech right now is getting your first job in tech." Comeau doesn't just quote it — he stress-tests it against the data. Indeed's job posting index for software development sits roughly 35% below its February 2022 peak as of Q1 2026, even as postings in nursing, skilled trades, and logistics have recovered to or above pre-pandemic baselines. Hacker News' "Who is Hiring" thread for May shows junior-tagged roles at their lowest absolute count since the thread's 2011 inception. Pull the LinkedIn data and the picture sharpens further: postings explicitly labeled "junior," "entry-level," or "new grad" are down approximately 50% from 2022, while staff and principal roles are down only ~15%.
Comeau's central claim is that this isn't a normal cyclical contraction. It's a structural reallocation: the work that used to justify hiring a junior — the small bug tickets, the boilerplate CRUD, the 'go read the docs and write a proof of concept' tasks — has been absorbed by Copilot, Cursor, and Claude Code running inside a senior's editor. The senior keeps the job. The ticket gets closed in 20 minutes instead of two days. The junior who would have closed that ticket, learned three things doing it, and asked a useful question at standup, is never hired.
The industry has been telling itself a comforting story for two years: AI raises the floor, juniors with Copilot will be as productive as mids without it, the ladder gets compressed but everyone still climbs. Comeau's essay is the most rigorous public takedown of that story to date, and he writes it from the position of someone whose business depends on it being true.
The key insight isn't that AI replaces juniors — it's that AI replaces the economic rationale for hiring them. A senior engineer in 2021 had a binary choice: do the grunt work yourself (slow, demoralizing) or hand it to a junior (fast for you, slow overall, but you build a pipeline). In 2026, the senior has a third option that wasn't on the table: hand it to an LLM, get it back in minutes, review it in ten. The junior was never competing with the senior. They were competing with the senior's tolerance for tedium, and that tolerance just got infinite.
Comeau quotes a hiring manager at a mid-sized SaaS company anonymously: "I used to hire two juniors a year because that was the only way to scale our team without burning out the seniors. Now my seniors aren't burning out. They're shipping more than they ever did. I haven't opened a junior req in 18 months and I can't justify one to finance." This is the mechanism. Not 'AI is better than juniors.' 'AI made juniors unnecessary for the specific economic role they used to play.'
The Hacker News thread split predictably. One camp — heavily represented by FAANG-tenured ICs — argued the market is just correcting from ZIRP-era over-hiring and juniors will be fine once rates come down. The other camp, mostly people who graduated bootcamps or CS programs in 2023-2025, posted application counts in the four digits with single-digit callback rates. The data favors the second camp. ZIRP correction would compress all levels roughly equally; the actual contraction is concentrated almost entirely at the bottom of the seniority distribution.
The deeper concern Comeau raises, and the reason this essay is going to be cited for years, is the demographic time bomb. Senior engineers were once juniors. The mid-level cohort that companies will desperately need in 2030 is being not-hired right now. Nobody has a plan for where 2030's senior engineers come from if 2026's juniors never got to write production code. Comeau notes — correctly — that mentorship and reps are not substitutable. You don't become a senior by reading Copilot's output. You become a senior by writing bad code, breaking production, and having someone more experienced explain why what you did was bad. Copilot can write the code. It can't be the person whose pager went off because of you.
If you're a senior or staff engineer, the implication is uncomfortable but actionable: you are now the bottleneck for your industry's future supply of senior engineers. The companies that figure out how to deliberately hire and grow juniors in an AI-saturated workflow will have a 5-year hiring advantage over the ones who let Copilot quietly eat their pipeline. Concretely, this means carving out work that an LLM *could* do but that you deliberately give to a human because the human needs the reps. It's economically irrational at the ticket level and strategically essential at the team level.
If you're hiring, audit your last 12 months of requisitions. If your senior:junior ratio has drifted from something like 3:1 to 8:1 without anyone explicitly deciding that, you're living inside Comeau's thesis. The fix isn't a token junior hire; it's restructuring sprints so juniors have protected work that compounds into expertise rather than competing with the LLM for the easy tickets.
If you're a junior or aspiring developer reading this: the path through is narrower but it exists, and it runs through demonstrating senior-shaped judgment earlier than juniors used to have to. That means open source contributions where you've argued architectural tradeoffs in PRs, side projects that show system design, and — Comeau's specific advice — getting genuinely good at code review, because code review is the one skill that scales up rather than getting automated away. The juniors getting hired in 2026 are the ones who look like mids on paper. That's a brutal bar, and Comeau doesn't pretend otherwise.
Comeau ends the essay without a tidy resolution, which is the right call. He doesn't know how this plays out, and neither does anyone else. What's clear is that the industry's two-year experiment in "raise the floor, everyone wins" is producing a different result than promised: the floor stayed where it was, and the ladder lost its bottom three rungs. The companies and senior engineers who recognize this as a collective action problem — and treat junior hiring as infrastructure investment rather than headcount expense — will be the ones with a functioning engineering org in 2030. Everyone else will be paying staff-level salaries to argue with Copilot.
The more time I spend accelerating my work with AI tools the more I realize how incredibly hard the craft of shipping useful software actually is.Sure, Claude Code and Codex can write (most of) the code for me - but the amount of technical knowledge I need to decide what and how to build remains eno
> I think AI tools are more like Iron Man's suit.There's an interesting repository with 63600 stars on GitHub (1). The developer of the repository is No 1 at the GitHub's trending contributors list (2). However, it seems like the application isn't what it's described to b
I also "had an Iron Man moment last week" as a mathematician. I've been doing joint math research with two friends (professors) on a project for several years. Last week, I decided to explore part of our research using chat GPT. 1) I would have a thought. 2) Present it to GPT. 3) Ask
I think that it's not a multiplier on skills.It's a reducer of time.For less experienced developers, it's an immediately reduction at the start of a project. But then they will almost certainly have problems later when their initial decisions come back to haunt them.For senior devs, i
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
I had an Iron Man moment last week where I was “vibe coding” a UI design with component tests live on the other screen. Iterating by asking it to move things, reduce emphasis of an element, exploring layout options, etc. The loop was near realtime and felt amazing.The code it generated was awful. Th