"vibe coding" was Collins' Word of the Year for 2025.

and the most popular take since: junior developers are finished.

honestly, the evidence says something more awkward. juniors aren't obsolete. but the WAY a lot of them use AI might make them so.

The consensus: vibe coding makes junior developers obsolete

the argument goes:

→ AI now writes code from plain English. that's vibe coding, which Collins defines as using AI, prompted in natural language, to help write code → juniors mostly wrote the simple code → so companies hire seniors plus AI and skip the juniors

and there's real data behind the worry. Stanford's Digital Economy Lab found that 22 to 25 year olds in the most AI-exposed jobs saw a 13% relative fall in employment through July 2025, rising to about 16% by October. young software developers saw one of the sharpest drops.

What the evidence actually says about junior developers

look closer though:

  1. the timing is shakier than the headlines. in a follow-up published this month, the same Stanford researchers say that with stricter controls the decline only becomes significant in 2024, and the earlier drops are probably at least partly down to other factors, not just AI
  2. AI isn't the clear speed boost people assume. in METR's 2025 study, 16 experienced open-source developers took 19% LONGER on real tasks with AI tools, whilst still believing they'd been about 20% faster
  3. the pros don't fully trust the output. in the Stack Overflow 2025 survey, 84% use or plan to use AI tools, but 46% distrust its accuracy and 66% say their top frustration is answers that are almost right, but not quite
  4. vibe coding isn't the job. about 77% of respondents in that same survey said vibe coding is not part of their professional work
  5. the people running the clouds want juniors. AWS CEO Matt Garman called replacing junior staff with AI "the dumbest thing I've ever heard", asking who'll know anything in ten years if nobody junior gets to learn. in December, Fortune reported him still warning that companies with no talent pipeline will eventually blow up on themselves

so the consensus gets the headline wrong. someone still has to review, debug and own all that "almost right" code.

Where the consensus has a point: the junior developer skills gap

here's the part worth taking seriously.

in January, Anthropic published a randomised trial with 52 mostly junior engineers learning a new Python library:

→ the AI group scored 50% on a follow-up quiz, against 67% for those who coded by hand → the AI group was only about two minutes faster, which wasn't statistically significant → the biggest gap was on debugging questions → people who handed everything to the AI averaged under 40%. people who asked conceptual "why" questions scored 65% or more

bad: AI writes it, you paste it, you learn nothing good: AI writes it, you interrogate it, you learn faster

the threat to juniors isn't AI doing the work. it's juniors skipping the part that turns them into seniors.

What a CS degree won't teach you about AI coding

most degrees still mark you on writing code from scratch. the job increasingly rewards:

  1. reading code you didn't write (AI or human)
  2. debugging fast, which is exactly where the AI group fell behind
  3. explaining why a working solution is still the wrong one

it's the same pattern we found across university more broadly in how students actually use AI: heavy use, very little training in judging the output.

The critical-thinking move for junior developers

five questions for every AI-assisted change:

  1. could I explain this diff line by line? (a diff is the set of changes you're proposing.) if not, it isn't ready, whatever the tests say
  2. what breaks at the edges? empty inputs, huge inputs, a dropped connection. ask the AI, then check yourself
  3. am I delegating or learning? in the Anthropic trial, asking "why" was the difference between 65% and under 40%
  4. is it actually faster? time yourself. METR's developers felt quicker and weren't
  5. who owns this at 2am? if it breaks in production, it's your name on it, not the model's

What to do this month

  1. debug without AI once a week. pick a real bug, solve it by hand, then compare with what the AI suggests
  2. review an AI pull request like a senior would. write down three things you'd change and why
  3. build one small project you can explain end to end, ready to walk an interviewer through every decision
  4. read the Anthropic study summary and pick the interaction pattern you want to copy

we'll keep tracking the field on our software engineering hub, and our guide to thinking critically about AI has the wider method.

so TLDR: vibe coding isn't making junior developers obsolete, but vibe LEARNING might, so use AI to understand faster, not to skip the understanding.

what's one bug you'd refuse to let an AI fix for you?

Sources

Get the next issue by email

theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.