What happens to the junior developer?

What the paper argues

If AI agents can write the code, is it still worth becoming a software developer? The answer is yes. The US Bureau of Labor Statistics projects software developer employment to grow 15 per cent between 2024 and 2034, against an all-occupation average of three per cent. What has changed is the route in. Junior developers used to build fluency through boilerplate, glue, tests and simple features. Agents now produce that volume in seconds, so the role shifts from writing code to reviewing and verifying it: less typing, more reading, more deciding whether the code in front of you is fit for purpose.

That is harder than what came before. It also means the advantages that now matter more, judgement, communication, system thinking and curiosity, are not locked behind a decade of experience. Anyone can start cultivating them today.

Key points

• The perception gap is the most important number. In METR’s mid-2025 randomised controlled trial, sixteen experienced developers predicted AI would speed them up by 24 per cent, estimated afterwards that it had sped them up by 20 per cent, and were in fact slowed down by 19 per cent.
• AI tools feel productive even when they are not. Stack Overflow’s 2025 survey found 84 per cent of developers use or plan to use AI tools, but only 29 per cent trust their accuracy. Veracode found 45 per cent of AI-generated code introduced vulnerabilities aligned with the OWASP Top 10.
• The foundations do not change: a programming language you can think in, Git, the terminal, SQL, HTTP and APIs, one framework learned properly, testing, and reading other people’s code.
• Beyond the foundations, the distinguishing layer is judgement: code review craft, working with agents, governance literacy, domain thinking and communication.
• The path in is harder. HackerRank’s 2025 report found 74 per cent of developers say finding a job is difficult, with junior hiring lagging. Companies are more interested in evidence: build something and ship it, contribute to open source, read pull requests in real projects.
• The traps include using agents to skip the foundations, imitating without understanding, and approving a pull request you do not really understand, which corrodes trust.