Two headlines from October sound like they contradict each other:

  • AI writes more code than ever. GitHub says one in three pull requests now involves an AI agent.
  • US software job postings are up nearly 15% since early 2025 (Indeed, via CNBC).

The detail explains it. 71% of that increase is senior roles. 37% of it is jobs with "AI" in the title. And Stanford's 2026 AI Index found employment for early-career software developers down 20% since generative AI took off.

The market is hiring people who can direct and check work. It's hiring fewer people whose main job is producing it.

What changed for juniors

The classic junior job was "take a well-defined ticket and implement it". That's exactly what agents are now good at. So the first step on the ladder got shorter.

The skills that are in demand are the ones that used to come later:

  • turning a vague request into a spec someone can build from
  • telling a correct change from a plausible one
  • knowing where the system actually breaks: money, concurrency, deploys, data
  • explaining a trade-off in writing

The good news: you can practise all of those now, without anyone hiring you first, and AI makes it faster to do.

What I'd build

Not another to-do app. Something with one hard problem in it, built to show that you understand that problem.

A few that work well, because each has a failure an interviewer can ask about:

  1. A booking system that can't double-book under concurrent requests. Show the race, then show the database constraint that ends it.
  2. Anything with money. Integer amounts, a ledger you can't edit, splits that never lose a paisa.
  3. A worker that processes messages and survives retries and duplicates without doing anything twice.
  4. An LLM feature with an eval set. Not "it works on my examples" but a number that goes up or down.

Then write about it. The blog post isn't marketing. It's evidence that you can explain a decision, and that's the senior skill.

How to use AI without hiding behind it

Use it for everything, but be able to answer these for any line in your project:

  • Why is this here?
  • What happens if it fails?
  • What test proves it works?

If you can't, the AI wrote it and you shipped it. Interviewers can tell, because they're seeing a lot of AI-built portfolios right now. The candidates who stand out can explain what they rejected.

What I'd tell you before you start

The entry-level door is narrower, not closed. It now opens for people who can show judgement, not just output. So build something where the judgement is visible, and write down every decision.

Sources: CNBC, "AI is automating coding. So why are software developer job postings on the rise?" (Oct 11, 2026), citing Indeed, GitHub and Stanford's 2026 AI Index.