Source: [Everyone Can Pass Your Coding Test Now. Here Is What to Hire For Instead](https://shrinath.me/blog/hiring-engineers-in-the-ai-era/)

# Everyone Can Pass Your Coding Test Now. Here Is What to Hire For Instead

When a candidate can generate a flawless answer in seconds, the take-home assignment and the algorithm quiz tell you almost nothing. What to look for in engineers now, how to run an interview that AI cannot fake, and the red flags worth more than any score.

By [Shrinath Prabhu](https://shrinath.me/), Senior Staff Engineer at Avail  November 12, 2026  5 min read

Picture a founder hiring their second engineer. They send the same take-home assignment they used two years ago to forty candidates. Thirty-eight come back excellent. Clean code, sensible structure, tests, a tidy README.

They cannot tell the candidates apart. Many of the submissions are nearly identical, down to the variable names.

The test did not get easier. It became irrelevant, because every candidate has access to the same very capable assistant. The assignment is now measuring whether someone could paste a prompt. That is a skill, but not one that needs a salary.

If you are hiring engineers now, or about to make your first technical hire, the old filters are broken. Here is what I would replace them with.

## What the old interview measured

The standard process tested three things: can you write code from memory, can you solve a puzzle under pressure, and do you know a framework’s details.

Those were always imperfect stand-ins for “will this person do good work”. Now they are not even that. Syntax, puzzles and framework trivia are exactly what AI is best at. Testing for them is like testing an accountant on long division.

## What to hire for now

The things worth paying for are the ones AI does not supply. I have argued elsewhere that [AI replaces coders, not engineers](https://shrinath.me/blog/ai-will-replace-coders-not-software-engineers/). Hiring is where that distinction costs or saves you real money.

**Judgment.** Can they look at a working solution and tell whether it is a good one? Most generated code runs. The question is whether it belongs.

**Problem framing.** Given something vague, do they start typing, or do they ask what you are trying to achieve?

**Ownership.** Have they stayed with something after launch, through the bugs and the boring parts?

**Communication.** Can they explain a technical trade-off to someone who is not technical? At a startup, that someone is you.

**Taste.** Do they notice when something is confusing, slow or half-finished?

**Honesty.** Do they say “I do not know” and “I think that is a mistake”?

## An interview AI cannot fake

**Let them use AI.** Banning it tests a way of working that no longer exists. Watching someone use it tells you a great deal: do they read what it produces, question it, catch its mistakes, or accept the first answer?

**Give a vague problem on purpose.** “Users are complaining the dashboard is slow. What do you do?” The strong candidate asks questions. Slow for whom? Since when? How do we know? The weak one starts optimizing something at random.

**Review code together.** Show them a pull request with a subtle problem. Not a syntax error, a design one: it works today and will hurt in six months. This is the job now, since much of an engineer’s day is judging code they did not write.

**Go deep on something they built.** Pick a project and keep asking why. Why that approach? What did you consider and reject? What went wrong? What would you change? Real experience has texture: surprises, regrets, specific numbers. Borrowed experience runs out of detail by the third question.

**Ask about a disagreement.** “Tell me about a time you thought your manager was wrong.” You learn whether they speak up, and how. I want people who will [tell me when I am wrong](https://shrinath.me/blog/how-to-disagree-with-leadership/).

**Change the requirements halfway.** Real work does this constantly. See whether they adapt or get rattled.

**Pay for a trial if you can.** A few paid days on a real, small problem tells you more than five interview rounds. It also lets the candidate assess you, which matters just as much.

## Red flags I take seriously

-   **Cannot explain their own submission.** The clearest sign there is.
-   **Never asks why.** Takes every requirement as given.
-   **No scars.** Everything they built went perfectly. Nobody’s has. Mine include almost nine years of saying “it works on my machine” with a straight face.
-   **Tool worship.** Describes themselves entirely by a framework.
-   **Blames others** for every project that went badly.
-   **Agrees with everything you say.** Pleasant in an interview, expensive afterward.

## Green flags

-   Asks about your users and your business before the tech stack
-   Says “it depends”, and then says on what
-   Volunteers a mistake and what they changed afterward
-   Pushes back on something in the interview, politely
-   Simplifies the problem before solving it
-   Has opinions and holds them loosely

## For founders hiring the first engineer

This hire sets your technical culture for years, so a few specific points.

-   **Hire for range over depth.** Your first engineer will do frontend, backend, deployment, support and some product thinking. A specialist is a later hire.
-   **Look for shipped things.** A finished, slightly ugly product beats a beautiful unfinished one.
-   **If you are not technical, borrow someone who is** for one conversation. An experienced engineer can tell in an hour what you cannot tell in five.
-   **Check that they can explain things to you.** If you leave the interview confused, you will be confused for the next two years.
-   **Do not hire the person who promises everything is easy.** Hire the one who tells you what is hard.

## On juniors

There is a temptation to stop hiring juniors because AI does junior work. I think that is short-sighted. Seniors come from somewhere.

What changes is what you look for. Not how much they already know, which matters less each year, but how fast they learn, how curious they are, and whether they insist on understanding what they submit. A junior who uses AI to learn faster is a great hire. One who uses it to avoid learning is not.

## The short version

Stop testing what a machine does better. Test for judgment, framing, ownership and honesty, in a setting that looks like real work, with the real tools.

The best signal in any interview I run is simple. At some point the candidate stops trying to impress me and starts actually thinking about the problem. I hire that person. It is the same quality that defines [what a staff engineer does](https://shrinath.me/blog/what-a-staff-engineer-actually-does/), only earlier in its development.

## Questions people ask

### How do you interview software engineers when they can use AI?

Let them use it, and assess what AI cannot provide: how they break down an unclear problem, the questions they ask, how they judge and correct generated code, how they explain trade-offs, and how they have handled real projects and failures in the past.

### Are coding tests still useful for hiring?

Traditional algorithm puzzles and unsupervised take-home tests have lost most of their value, because AI solves them instantly. A live working session on a realistic problem, with AI tools allowed, is far more informative.

### What should a founder look for in a first engineering hire?

Ownership and judgment over raw coding speed. Someone who asks why before how, has shipped and maintained something real, can explain decisions in plain language, and will tell you when they think you are wrong.

---

[Agent docs](https://shrinath.me/agents/index.md) · [llms.txt](https://shrinath.me/llms.txt) · [All pages](https://shrinath.me/sitemap.xml)
