A take-home assignment tells you less on its own now that an artificial intelligence (AI) tool can write most coding exercises in minutes. It’s a short piece of work a candidate completes on their own time and then talks through with the team, and it still has a place. Use one when the role needs design judgment that a timed live coding round squeezes out, cap it at a few hours, let candidates use AI and ask them to say how, and score the conversation about the work rather than the submission alone.
Teams often choose between a take-home and live coding on logistics, such as which is easier to schedule. That choice now needs more thought, because two candidates can hand in code that looks the same. What still separates them is whether they can explain it, change it, and say where the AI tool went wrong.
I recommend never scoring a take-home assignment on the submission alone: the submission earns the candidate a follow-up conversation, and the conversation is what you score.
This guide is for recruiters, hiring managers, and engineers who run technical interviews. It covers when to use a take-home and when to use live coding, how to design and score a take-home, how to run a live round that tests reasoning, and how to keep the evidence for the debrief.
Choose a take-home assignment or live coding for the role.
Both formats are work samples, and each suits some roles better than others, so choose by the role in front of you. Use this table as a starting point and change it for your team:
| Situation | Take-home assignment | Live coding |
|---|---|---|
| Senior and staff engineers | Strong: design judgment needs more room than a short round gives | Best inside the follow-up conversation, as a live change to the take-home |
| Early-career hires | Weaker: with less experience to draw on, their submissions tend to look alike | Strong: the reasoning is what separates candidates |
| Roles where most candidates hold other offers | Costly: unpaid hours are easy for them to turn down | Strong: one scheduled interview |
| Pairing and mentoring roles | Weak: you never see them work with anyone | Strong: working together is what’s being tested |
| Roles where strong people often interview badly when watched | Strong: this is who take-homes are for | Risky as the only test: pair it with something else |
Whichever format you choose, decide before the role opens which skills it tests, and use the same format for every candidate at that stage. Switching formats mid-search makes candidates hard to compare.
Design a take-home the candidate can finish in a few hours.
A take-home is the part of your process that looks most like the job: real tools, time to think, and no audience. Design it so it stays that way, and so the follow-up conversation has something to work with:
- Cap it at a few hours, and have an engineer on the team do it first to check it fits. Tell candidates you’ll judge it as a few hours of work and won’t score anything past the limit. A long take-home favors people with free evenings.
- Build it on a small sample codebase. Ask for a change to code the candidate didn’t write, so they have to read and understand it before changing it, as they would on the job. Add one constraint that rules out the obvious approach, so the candidate has to make a design choice they can explain in the follow-up conversation.
- Allow AI tools, and ask for a short note on what the candidate used and where they changed what the tool produced. You can’t enforce a ban, and a ban rewards candidates who hide what they used.
- Tell candidates what you’ll score and that a conversation about the work follows.
- If the assignment has to run past a few hours, or the work touches your real product, pay for it.
Designed this way, a take-home and the conversation about it are proof of work: evidence of what the candidate can do, rather than a line on a profile. On the 10x Recruiting podcast, Jon Bischke talks about why proof of work beats LinkedIn profiles. Bischke founded Entelo and now leads ZoomInfo’s recruiting product.
Score the conversation about the work.
The follow-up conversation is where a take-home earns its score. Book a follow-up call with the candidate and an engineer who has read the submission, and run it in this order:
- Ask the candidate to walk through the choices they made, and why.
- Ask how they used AI: what they asked it, what came back wrong, and what they threw away.
- Ask for a change they didn’t plan for, such as a new requirement, a failing edge case, or a performance limit.
- Score what they said and did against the skills you set before the role opened.
A candidate who understands their submission can reason through a new requirement out loud, even if they don’t finish the change. A candidate who doesn’t understand it tends to stall on the first change that touches their own design.
For a fuller way to assess how engineers build software with AI tools, read the guide to assessing the vibe-coding competency.
Run live coding so it tests reasoning.
Live coding shows what a take-home can’t: the reasoning as it happens. You see what a candidate does when they’re stuck, whether they ask before assuming, and how they take a correction. Set the round up so it tests that, and not how calm someone stays with an audience:
- Let candidates use their own tools: an editor, documentation, and an AI assistant. Then watch how they direct the AI assistant.
- Pick a problem that leaves something unclear on purpose, and give the first few minutes to the candidate’s questions before any code is written.
- Talk less than the candidate. Ask a question, then let them think out loud.
- Debug something together. Fixing broken code with the candidate is closer to the job than solving a problem from scratch.
A good interviewer works through the problem with the candidate. Shahriar Tajbakhsh, Metaview’s co-founder and chief technology officer, put the difference between a good and a bad interviewer in two sentences:
A bad interviewer transacts. A good interviewer interacts.”
For the problem itself, the hints to give, and how to score each stage, there’s a guide on how to lead a coding interview.
Write the evidence down before the debrief.
A well-run technical round still leaves the panel guessing by the debrief if no one wrote down what happened. The evidence from a live coding round or a follow-up call sits in the interviewer’s head and fades into an impression. Keep it with these steps:
- Write the scorecard the same day, while the code and the conversation are fresh.
- Record what the candidate said and did for each skill, in their words where you can, before the overall recommendation.
- Note what you didn’t get to test, so the next interviewer or the debrief can pick it up.
- When the panel disagrees, go back to what each interviewer saw. A split often means two people watched different parts of the round.
A scorecard that records what the candidate did lets the panel compare candidates on the same skills, whichever format the role uses.
Where Metaview fits.
Metaview is the Agentic Recruiting Platform. In a technical hiring process, it helps with the evidence this guide asks you to keep: a record of the conversation about the work.
Where consent has been given, the Metaview Notetaker records the follow-up conversation or the live coding round and turns it into structured notes. The recording captures every spoken word, so the scorecard starts from what the candidate said about their choices rather than from memory.
The Notetaker then drafts the scorecard from the conversation. The interviewer reviews the draft, rates the candidate, and submits it, so the recommendation stays theirs.
How to tell your take-home assignment is working.
After the first few candidates, use these five checks to see whether the take-home and the conversation about it are telling you what you need:
- Do candidates finish within the time limit? Ask each one how long it took, and if most go over, cut the scope.
- Do the follow-up conversations differ from one candidate to the next? Similar submissions are expected now. If every candidate also explains and changes their work equally well, make the unplanned change harder, or use live coding the next time you hire for this role.
- Does every follow-up conversation include a change the candidate didn’t plan for? If not, add one to the interview plan.
- Are candidates with other offers dropping out at the take-home? If so, pay for the work, or use live coding the next time you hire for this role.
- Can the panel explain each decision from the scorecards? If not, write them the same day, with what the candidate said and did for each skill.
A take-home assignment works when the submission leads to a conversation about the work, and that conversation is what you score. Keep it short, let candidates use the tools they’d use on the job, and score what they explain.
Keep the evidence from every take-home conversation.
See how Metaview records the conversation about the work and drafts each scorecard for the interviewer to review and submit.
Frequently asked.
Is a take-home or a live coding interview better for hiring engineers?
Neither is better for every role. A take-home suits senior work where design judgment needs room, and live coding suits early-career hires and roles built on pairing. In both, the conversation about the work tells you more than the code alone.
How long should a take-home assignment be?
Keep it to a few hours, and tell candidates you won’t score work done past the limit. A longer assignment favors candidates with spare time and loses those who are interviewing elsewhere.
Should candidates be paid for take-home assignments?
Pay for it if the assignment has to run past a few hours, or the work touches your real product. A short, capped exercise on a sample codebase is reasonable to ask for unpaid.
Should candidates be allowed to use AI on a take-home?
Yes. You can’t enforce a ban, and the job is likely to involve AI tools. Ask for a short note on how the candidate used them, and talk it through in the follow-up conversation.
How do you tell whether a candidate understands the code they submitted?
Ask them to change it live for a requirement they didn’t plan for, such as a new edge case or a limit that breaks their current structure. A candidate who wrote and understood the code can talk through how they’d adapt it, even if they don’t finish.