Most engineers I interview don't write their code by hand anymore. Claude Code writes it. Or Copilot, or another AI tool.

That raises a question few hiring teams want to ask out loud. Do developers still need to know how to code?

What I Keep Seeing in Interviews

Over the past couple of months I've noticed a pattern. I ask engineers how they work day to day, and the answer is almost always the same. The AI writes the code. They prompt it, read what comes back and fix it.

Many of them told me it had been a while since they last wrote a line of code by themselves.

So I asked some of them to do it. I give candidates a very simple coding task that a first-year computer science student could finish. It needs no external libraries and runs in an online compiler if it has to. It asks for a data structure and a loop or two, and nothing more. Candidates can ask me for help and search for syntax. The only rule is that they don't use AI-generated help.

Then I sit and watch. A larger share of candidates than I expected can't finish it. A few were honestly embarrassed by how slow they were.

My first thought was that this might only be my small sample.

The Numbers Behind It

It isn't. JetBrains asked more than 15,000 professional developers earlier this year how much of their code they write by hand. Here's what they found:

  • On average, about 47% of their code was written entirely by AI agents.
  • More than half wrote less than 20% of their code by hand.
  • About one in five wrote no code at all without AI help.

Only about 23% still write most of their code themselves. JetBrains' own conclusion is that purely manual coding is quickly becoming a thing of the past.

The Stack Overflow 2026 Developer Survey came out in October with more than 30,000 respondents. Of the people who use AI coding tools, 73% use them every day. Claude Code is now the most used coding agent.

So the engineers in my interviews aren't the odd ones out. They are the normal ones.

Why It Matters Who Writes the Code

If the AI writes the code, why should anyone care whether you can?

Writing code was always about more than typing. It was how engineers learned to think through a problem step by step. Early research suggests that when we hand the writing over, some of the thinking goes with it.

Engineers learning a new library

Anthropic ran a randomised trial published in January 2026. 52 mostly junior engineers learned a new Python library. Half of them used an AI assistant and half didn't. Then everyone took a quiz.

The AI group scored 50% and the group without AI scored 67%. The biggest gap was in debugging, which is exactly the skill you need when the AI gets it wrong. The AI group wasn't much faster either.

The detail I like most is that the way people used AI mattered. Those who let the AI write everything scored below 40%. Those who used it to ask questions and understand the code scored 65% or more.

Knowledge workers trusting the answer

Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real tasks they did with AI. The more people trusted the AI, the less critical thinking they reported. The more they trusted themselves, the more they checked, questioned and corrected the answer.

Students writing essays

In a small study at the MIT Media Lab, 54 students wrote essays with ChatGPT, with a search engine or with no tools at all. The ChatGPT group showed the weakest brain connectivity. Most of them couldn't quote a single line from the essay they had just written.

It's an early preprint that hasn't been peer reviewed yet, so treat it as a signal. It points the same way as the other two studies.

What This Doesn't Mean

None of these studies says AI makes engineers worse at their jobs. The JetBrains numbers show that this is simply how the work gets done now. A hiring process that pretends otherwise is hiring for a job that no longer exists.

What the research suggests is narrower. If you stop doing the thinking, you slowly get worse at it.

My co-founder Todor often says, "A fool with a tool is still a fool."

AI makes a good engineer faster. It doesn't make a weak engineer good. It helps them produce more code they don't understand, and produce it faster.

So do developers still need to type code from memory? Maybe not. Do they need to read it, judge it and fix it when it breaks at 2am? More than ever. That changes what a technical interview should test.

Three Things to Change in Your Next Interview

Here's what we've learned actually works, after running thousands of technical interviews:

  1. Test reading and debugging. Give candidates a short piece of code with a bug in it. AI-written code works well for this. Ask them to find the bug and explain it. That's where the Anthropic study found the biggest gap, and it's what the job looks like today.
  2. Keep 15–20 minutes without AI. You don't need a two-hour coding marathon. A short, simple problem that the candidate talks through out loud is enough to see whether they can reason on their own. Tell candidates in advance which part allows AI and which part doesn't. Don't penalise slow typing, because most engineers are slow without AI now. Being slow is fine. Being lost is the problem.
  3. Watch how they use the tool and score the reasoning. In the part where AI is allowed, notice whether the candidate asks the AI questions or pastes its output and hopes for the best. Ask "why" after every AI-assisted answer. Put the reasoning on your scorecard and leave speed off it.

This is an afternoon of work. Possibly less.

So, honestly, how little should someone be able to code on their own today and still be called a developer? And in your interviews, do you still ask engineers to code without AI?

Sources

This post first appeared in Shortlisted by Kat. Photo by Alicia Christin Gerald on Unsplash.

Interview for Thinking

Every AssessDev tech assessment ends with a live challenge of about 30 minutes without AI tools. Candidates can ask questions and search for syntax, and we score how they reason.

Tech Assessments

Structured technical interviews run by senior engineers and CTOs, with a live challenge and a report within 24 hours.

Tech Interviewer Training

Eight hours of live training that helps your engineers and hiring managers run structured interviews in-house.