I have been tracking junior software-engineering listings on LinkedIn and Indeed since late 2024. The graph has one shape. Flat in 2024, sloped down through 2025, falling off a cliff through Q1 2026. Big tech is openly senior-only. Mid-tier has followed. The few junior roles that did go up in 2026 advertise lower salaries, higher experience floors, and a take-home test designed to detect LLM use rather than ability.

The data is consistent across sources. Junior developer postings are down around 40% compared to pre-2022 levels, and entry-level hiring at the fifteen biggest tech firms fell 25% from 2023 to 2024. Computer-science graduate unemployment hit 6.1–7% in 2025, the fifth-highest among US college majors. According to a recent LeadDev survey, 54% of engineering leaders plan to hire fewer juniors in 2026 because AI copilots let seniors absorb more work.

AI didn’t start the junior contraction. The post-2022 retraction did. But AI is the rationale boards lean on to keep it going. The apprenticeship pipeline that produces senior engineers has a 5–7 year lag. The cohort we aren’t hiring today is the senior bench we won’t have in 2031.

The predecessor to this piece argued AI didn’t break engineering. We did, by treating it as a quota multiplier. This piece is what that mistake looks like when you compound it five years out.

How Seniors Used to Get Made

Senior engineers aren’t produced by classroom learning. They’re produced by an apprenticeship of specific moments, and almost all of those moments are exactly what AI is removing.

The arc takes five to seven years on average. First commit to staff-level judgment isn’t a graduation timeline, it’s a feedback-loop timeline. You learn to be senior by getting your pull request torn apart in review and having to defend the design choice. By debugging a production incident at 2 a.m. with nobody to ask, when the LLM either doesn’t exist or doesn’t know your system. By owning a feature whose failure mode you didn’t anticipate, and writing the postmortem with your name at the top. By being the person who has to explain in a room of executives why the migration broke.

None of this transfers through documentation. It transfers through accountability. Your name on the broken deployment, your call in the heated design review. And AI risks shortcutting the moments where the cost of failure used to land on the engineer, not on the tool. That is where senior judgment gets built.

In the previous piece I catalogued four habits that hollow out code review: copy-paste without reading, stack traces becoming LLM input, AI-generated tests no one validates, pattern atrophy. Those four habits are also the four exact failures of the apprenticeship loop. Same problem at the per-developer scale of the previous piece, same problem at the industry scale of this one.

You can’t grow a staff engineer in a classroom. You can only grow one in a feedback loop. Apprenticeship is the most common one, self-directed practice with brutal review is the other. AI didn’t remove a job. It removed the apprenticeship loop that produced the next generation of seniors at scale.

What Hiring Actually Did in 2025–2026

This isn’t a prediction. Most of the damage landed in the eighteen months between late 2024 and the end of 2025, and you can feel it in the loops.

I ran two junior hiring loops in 2024 and one in 2025. In 2024 we worked from three good shortlists out of roughly forty applicants per role. In 2025 we had one passable shortlist out of two hundred. The pool wasn’t smaller. The floor was lower. Every recruiter and engineering manager I talked to that year had a version of the same story. We were fishing in the same shallow water, with the same complaints, and the candidates who did show up couldn’t pass a whiteboard problem they’d have solved in 2022.

The macro numbers track what the loops felt like. Junior developer postings are down roughly 40% compared to pre-2022 levels, and entry-level hiring at the fifteen biggest tech firms fell 25% from 2023 to 2024. Employment for software developers aged 22 to 25 has dropped nearly 20% from its late-2022 peak. The contraction concentrates at one end of the ladder. Senior listings stay roughly stable while junior listings fall, which is the shape that should be worrying anyone budgeting for a 2030 staff bench.

The Klarna announcement is the one I kept hearing back about. Roughly 700 jobs cut between 2022 and 2024, AI credited for the savings, the CEO doing a victory lap in interviews about how efficient they’d become. Salesforce and Meta followed through 2025 and into 2026 with their own AI-credited cuts. None of those announcements were the damage. The damage was that by mid-2025 I’d heard “we don’t need a junior team, AI is our junior team” verbatim from two leaders at two different companies, neither of which had announced cuts of their own. They’d read the press releases. The spreadsheet wasn’t necessary.

Hiring leaders didn’t help themselves on the way down. Cheating on technical take-homes more than doubled in six months, from around 15% in June 2025 to 35% by December 2025. Cluely and Interview Coder showed up in candidate Discords as named tools, not whispers. The hiring-side response was predictable: drop the take-home, raise the experience floor on the job spec, add a live conversational-AI interviewer that probes reasoning. Each of those responses cut juniors further. The take-home was the format juniors had the best shot at, and replacing it with a real-time live coding session is exactly the format juniors fail. I wrote about the candidate-side fix last year. The hiring-side fix is harder because it costs senior engineering hours the company already isn’t protecting.

The juniors who did get in are being trained differently from the cohort five years ahead of them. 56% of CHROs surveyed in 2026 said early-career talent turn to unsanctioned AI tools when formal guidance is unclear. 38% of leaders worry early-career hires aren’t building long-term skills like judgment, communication, and critical thinking. Watch a 2024-hired engineer in their first incident review. They’ve never written a regex without an LLM. They’ve never sat with a stack trace longer than it takes to paste it into a chat window. The predecessor put it bluntly. They learned to lean on AI because we gave them velocity targets, no mentorship budget, and an LLM subscription, and called it onboarding.

None of this started with a thinkpiece. It started in the headcount meeting where someone first said “we can absorb that with AI” and the room nodded.

The Mid-Level Trap

The bill is already coming due. Not in 2031. In the 2024–2026 cohort sitting at IC2.

Let me describe a composite character. AI-native from day one. Hired late 2023 or 2024 into a team running on velocity dashboards. Ships fine in normal flow. Features get out, the PR description reads cleanly, the tests are green. Can’t architect a new system from a blank page. Can’t debug a production incident without the LLM in the loop. Can’t mentor a more-junior teammate because the pattern library isn’t there to draw from.

Plenty of engineers hired in that window fought for the feedback loop on their own. Sought out mentors, took the unglamorous on-call shifts, asked the unfashionable questions in review. They’re the exception. The median outcome of the velocity quota the predecessor described is the deficit profile above. The cohort learned to ship under AI assist. The skill that didn’t get built is the one you can’t fake under pressure.

Burnout data tracks the same line. Harvard Business Review’s March 2026 study found 62% of associates and 61% of entry-level workers reporting burnout, with 83% saying AI increased their workload. The participants used the phrase “brain fry”. That fry concentrates in the 2–3 year cohort because the gap between what the dashboards say they can do and what they can actually do under pressure is widest there.

The structural point is what comes next. This cohort can’t be promoted to senior on the current trajectory. The bridge from mid to senior is exactly the skill set they haven’t had a chance to build. Judgment under uncertainty, novel debugging, system thinking, the willingness to be wrong in front of more-junior people. They’re stuck.

What that engineer should do is the question. Find a mentor and ask for explicit calibration on the skills not on the dashboard. Take on one project a quarter where AI doesn’t touch the hard part. Become the exception, not the median. The last section of this piece is what the org should schedule. This is the part that doesn’t wait for it.

The 2029 mid-bench is already weaker than the 2024 mid-bench. What the 2031 senior bench looks like is the question that follows.

The Five-Year Math

This isn’t a dip. It’s a structural collapse, and the math is dispassionate.

The current senior cohort has its own attrition. Turnover quietly steady at industry rates. The mid-2010s boom cohort is approaching the retirement window. None of this is dramatic, and none of it’s new.

The new thing is on the supply side. Companies are already reporting staff+ vacancies that take 66 days on average to fill, 50% longer than non-technical positions. Senior engineer hires through traditional channels run four to six months. The lemon.io 2026 survey called the shape of the talent market a “bifurcation problem”. The junior side flooded with applicants nobody is hiring, the senior side empty. The shortage is already, in 2026, exclusive to senior-level roles. The 2031 prediction isn’t “this will happen.” The early signs are in the 2026 fill-time data. Five more years of compounding is what comes next.

The model isn’t complicated. Top-15 firms cut entry-level hiring 25% in 2024. Junior postings overall fell roughly 40% by 2025–2026. Combine that with a 5–7 year apprenticeship lag and the 2030–2031 staff+ pipeline thins out rather than refills. The exact number of unhired juniors is debatable. The direction isn’t.

AI isn’t the only force behind the contraction. The post-2022 retraction and higher interest rates accelerated it. But AI is the lever organisations point to when defending the cut. Even if junior hiring fully reverses in 2027, the apprenticeship lag pushes the impact out to 2032–2034 rather than removing it.

A senior shortage doesn’t resolve by hiring harder. It resolves by one of three paths. Import seniors from other companies, which is zero-sum at the industry level. Promote mid-levels prematurely, which produces the postmortems you don’t want to read in 2032. Or wait five to seven years from the day junior hiring restarts, which means the only intervention that matters has to start now.

No quarterly OKR is going to drive that. It has to be a leadership decision against the quarterly clock.

”But the AI-Native Senior of 2031 Will Be Different”

The honest counter-argument is that future engineers will mature differently. AI-as-coworker from day one will produce a different kind of senior. Faster on the easy work, building judgment through a different feedback loop. Maybe the 5–7 year lag becomes a 2–3 year lag. The surface of “what a senior needs to know” has shifted before. IDEs, Stack Overflow, offshoring. Each one was supposed to break the next generation. Each time it didn’t.

The reason this time is different isn’t theory. It’s evidence.

Klarna already ran the small-scale version of this experiment (in customer service, not engineering) and rolled it back. By early 2026, Klarna’s CEO publicly admitted the AI-for-headcount swap had degraded service quality, and the company began rehiring. The lesson isn’t that AI-native engineers fail. It’s that boards swap human capacity for AI capacity faster than the data justifies, then overcorrect when the bill arrives. Salesforce, Meta, and Microsoft are running the same swap in functions much harder to roll back than customer support.

The deeper question is whether senior work is the kind of skill that can be built on an AI-mediated path at all. Sitting with a postmortem until the second-order failure no one named becomes visible. Knowing which of the seven retry layers is masking the real failure at 2 a.m., when the LLM can’t see your system’s history. Choosing the boring architecture in a room of engineers who all want the interesting one. None of that’s built by AI-assisted easy work. It’s built by being the person whose name is on the postmortem.

The AI-native senior of 2031 will exist, and they will be faster at the easy work. Whether they can do novel-problem work without the feedback loops that build that judgment is the open question, and the early signal is not encouraging. The engineering version of the same experiment is harder to abort than the customer-service version was. The quality drop won’t surface until the next cohort fails to graduate to senior. By the time the bill arrives, the rollback window has closed.

Hire the 2031 Senior in 2026

These are the basics a lot of organisations have stopped doing.

For VPs, CTOs, and heads of engineering. Reopen the junior pipeline now. The 2031 senior is hired in 2026. Set a junior-to-senior ratio. One junior per five senior engineers is a defensible floor. And put it in your annual plan as a locked headcount line, counter-signed by the CFO. Treat it the way you treat compliance headcount, not the way you treat the AI tooling budget. If the board asks why headcount is going up when AI is supposed to bring it down, point at the 66-day staff-engineer fill time already in your own recruiting data. The shortage isn’t a 2031 prediction. It’s a 2026 cost line that gets worse every quarter you delay.

For hiring managers. First-year juniors don’t get AI-velocity targets. The first twelve months of an engineering career are the apprenticeship. Fully, no exception. This isn’t a call you can make unilaterally at the team level; surface it upward, get it agreed at VP, and put the lower throughput in the plan before the dashboard does it for you. The team you have in 2031 is the one shaped by the targets you protected in 2026.

For tech leads and staff engineers. You’re the apprenticeship. Block two hours a week on the calendar for mentorship that survives throughput pressure. Mentorship that happens “if there is time” never happens. The 2031 staff engineer is built by what you choose to protect on the calendar in 2026.

For senior engineers. If nobody on your team can become you, you can’t leave the team. That’s your problem, not your employer’s. Train someone anyway. The optionality is yours to defend.

For early-career engineers and juniors. Pick the team that mentors. Walk past higher offers if the higher-paying team ships at AI velocity with no mentor cover. The senior career you want in 2031 is built by the manager who protects your first twelve months in 2026.

For solo founders, OSS maintainers, and anyone owning the whole stack. The apprenticeship loop you don’t have is the one you have to build. Pair with someone who’ll tear apart your design choices. Read postmortems until you can predict the second-order failure before it lands. The pipeline collapse doesn’t apply to you. The skill collapse does.

AI didn’t kill the senior engineer. We did, by not hiring the junior who would’ve become one. Compound interest works in both directions. The per-developer cost I wrote about a month ago is one side of the meter. The unhired junior is the other, and that meter has been running since 2024 whether the dashboards report it or not.