What replaces the coding interview? Four options and what each shows

Anand Chapla ·

Nothing has replaced it yet, and nobody has shown that any one option works better. But the options are not equal. The ones that show how a person works, with the tools they will really use, give you the most to look at. A short real task, done with an AI assistant, with the whole session kept on record, is the strongest of the four below. A portfolio, a code review round and a pair session each show one part of the picture.

Who stopped, and what they said

On 27 October 2026, Felix Rieseberg published “Hunger beats tenure: Advice for junior engineers” at felixrieseberg.com. He is an engineering manager at Anthropic, and he wrote: “This month, I laid the traditional coding interview to rest.” He also wrote: “As a hiring manager at Anthropic, I stopped asking candidates to do them.”

He did not name a replacement. He wrote: “I haven’t quite gotten there yet”. And he asked for help: “If you have figured out something better, I’d love to hear how.”

That is an honest place to be, and many hiring managers are in the same place. If you are one of them, you have an empty slot in your process and four common ideas to put in it.

Option 1: a portfolio review

Felix describes his own way as “scanning resumes and portfolios for little prototypes that exhibit taste, empathy, and principled thinking.”

What it shows: what a person has built, and what they care about. It costs the candidate no extra time, and it can show taste, which a puzzle never could.

What it cannot show: how they work. A finished project does not tell you which parts the person decided, which parts an assistant wrote, or what they did when something broke. It also favors people who had free time to build things, and many good engineers did not.

Option 2: a code review round

You give the candidate a change written by someone else, and ask what they think of it.

What it shows: reading skill and judgment. Reviewing a change is a large part of working with AI, because someone has to read what the assistant wrote.

What it cannot show: what they do when they have to start from nothing. Reading is not the same as building, and a person can review well and still struggle to ship.

Option 3: pair programming, with AI allowed

You sit with the candidate, they build something, and you talk while they do.

What it shows: how they think out loud, and how they work with another person. You see the work as it happens.

What it cannot show, reliably: the same thing twice. Two interviewers pair differently, and the score depends on who sat in the room. Each pair session also costs one engineer an hour or more, for every candidate. And what you remember an hour later is not a record.

Option 4: a short real task, with the same AI access they would have on the job

The candidate gets a small real problem, a time limit, and an AI assistant. Then you look at what they did.

What it shows: the most. How they split the task, what they asked the assistant, what they checked before they trusted an answer, what they kept and what they deleted, and whether the result works.

What it needs: a record. If the only thing you keep is the final code, you have thrown away most of the evidence. Without the record of the session, you are back to a gut feeling about a finished file.

What to ask of any option

Whichever you choose, ask three questions.

  1. Does it show the work, or only the result? The result is easy to get now. The way the person got there is the part that tells you about them.
  2. Can someone who was not in the room check the score? A good score points to a moment in the session, a line of code or a test. If it cannot point to one, it is an opinion.
  3. Can two interviewers reach the same answer? If the score changes with who is interviewing, it is measuring the interviewer.

What nobody knows yet

No study we know of shows that any of these four predicts job performance better than the old coding interview did. The old interview was never shown to predict it well either. So choose the format that gives you the most to look at, keep the record, and check your own hires against it after six months. That is the only evidence you will have, and it will be yours.


Our only outside source is the post quoted above, read at its own page on 10 November 2026. The four options and the three questions are our own argument, and we build a product based on the fourth option, so read this with that in mind.

codesolara gives every candidate a real VS Code workspace with an AI assistant next to them, and every score points at the moment behind it. For what to look at in that session, see how to assess an engineer who uses AI.

// more