Freeze Junior Hiring Now, Run Out of Seniors by 2029

The agent writes the feature in an afternoon. The person who can tell you it's wrong took 6 years to make.

The line goes something like this: "We're not hiring juniors anymore. Claude does that work." It gets said like a cost win, and on this quarter's spreadsheet it is one.

The whole industry is making the same bet. A Stanford Digital Economy Lab team tracking ADP payroll data found that employment of workers aged 22 to 25 in AI-exposed jobs now sits 19% below where it would otherwise be, and the gap comes mostly from companies hiring fewer of them (layoffs barely move it). Experienced workers show no comparable drop.

So the bottom rung of the ladder is getting sawn off. Quietly, one skipped req at a time.

Where seniors come from

Here's how I run engineering with AI: the 70/30 split. Agents do the mechanical 70% of the work, and a senior human owns the 30% that's judgment: what to build, what to reject, what will break at 3am.

That 30% is the whole point. And it only works if someone on your team can read an agent's pull request and smell the problem before it ships.

That nose takes years to grow. You get it by shipping bad code as a junior, watching it fail in production, and getting your work torn apart by someone better. Every senior engineer you'd want to hire in 2029 is a junior somebody hired in 2026.

The data says fewer of those hires are happening. SignalFire's talent report found that at the biggest tech companies, new grads are just 7% of hires, down over 50% from 2019. Startups aren't far behind: under 6% of hires, down over 30%.

And engineering leaders expect more of it. In LeadDev's 2025 AI Impact Report, 54% said AI coding tools would reduce junior hiring over the longer term. That's a survey of expectations, so treat it as mood. The mood is grim.

Run that forward 3 years. The seniors you have now get poached, burn out, or start their own companies. The pool you'd replace them from is half the size it should be. Everyone's bidding for the same people, and you're paying 2029 prices for judgment you could have grown in-house.

The people running the biggest AI bets disagree with you

AWS has more reason than almost anyone to sell you on agents replacing people. Its CEO, Matt Garman, called replacing junior developers with AI "one of the dumbest things I've ever heard," and said dropping junior hiring entirely is "a nonstarter for anyone who's trying to build a long-term company." His point is boring and correct. Juniors are usually your cheapest people, so cutting them first to keep pricier seniors makes no financial sense. "At some point that whole thing explodes on itself."

IBM went further. It plans to triple entry-level hiring in the US in 2026, and it rewrote the job descriptions first. Its junior developers spend less time on routine coding and more time with customers.

That rewrite is the part founders skip.

What a junior does now

The 2019 junior job was typing boilerplate. Agents own that now, and they're welcome to it. A 2026 junior is an operator and first-pass reviewer of agent output. In practice that looks like 4 things.

They run the agents on scoped work. Tests, migrations, small features with a clear spec. They write the prompt, check the result, and own the ticket until it's merged.

They do the first review. The junior reads every agent PR before your senior does and flags anything they don't understand. Half the time the thing they don't understand is the bug.

They write down what broke. Every agent mistake goes into a running log: the prompt, the output, the fix. In 6 months that log is your team's real playbook.

They sit next to customers. Support tickets, sales calls, bug reproductions. Agents can't do this part, and it's where juniors learn what the code is for.

Your senior's job changes too. They review the junior's review, and they say out loud why they'd reject something. That narration is the training program. It costs maybe 3 hours a week.

Here's how it usually plays out. A Series A team freezes junior hiring and leans on 2 seniors plus a stack of agents. Output goes up for 2 quarters. Then one senior takes a bigger offer, and the whole review queue lands on the one who's left. Agent PRs start merging after a skim instead of a read, because there's no one else to read them. The bugs surface about 3 months later, usually in billing.

When not hiring one is right

Sometimes skipping the hire is the right call. I see 4 cases.

You have no senior to train them. A junior with only agents for mentors learns the agents' habits, including the bad ones. Hire the senior first.

You have under 12 months of runway. Training pays off in year 2. If you might not have a year 2, spend the money on shipping.

Your product is mostly glue. If the codebase is a thin layer over 3 SaaS APIs, there may not be enough judgment work to grow anyone. Revisit when that changes.

You're about to rebuild. Hiring someone to learn a system you're throwing away in 4 months wastes both of you.

My rule of thumb is 2 questions. Will a senior on the team spend 3 hours a week teaching? Can you carry the salary for 18 months? If both answers are yes, hire the junior. If either is no, wait.

Outside those cases, hire one. One is enough for most seed and Series A teams. Give them the agents, a senior who explains things, and a customer to talk to.

The agent will keep writing the code for free. You'll pay for the person who can read it in 2029, whether you start growing them today or not.