In 2020, running design at Grammarly, I stopped sending assignments as part of the interview loop. I have not missed them once, and the arguments I hear for keeping them have not improved in the six years since.
This was on a design team of several dozen, with a hiring load that made the loop visible. Volume is what forced the question. When you run four candidates a month you can absorb a bad stage in the loop. When you run four a week you find out exactly what each stage in your loop is measuring, because you see the same failure repeatedly. By 2021 our director loop was an hour with two or three projects, in the room, nothing to take home. Six years earlier, hiring at Honor, I had set the exercise as a take-home myself. I would not do it now.
What the assignment measures
Its stated purpose is to see how someone thinks about a problem. In most companies its real purpose is to produce something a hiring committee can look at. Looking at a piece of work feels more objective than making a judgement about a person.
What it measures in practice is spare time.
A strong candidate without dependents, currently unemployed or in a job with slack in it, gives you fourteen hours polished to a very high finish. An equally strong candidate with two small children and a demanding role gives you four hours. It looks worse. Your committee reads the difference as ability. I have watched that comparison happen in a room and I have made it myself. Nobody in that room is trying to filter for household circumstance. They filter for it anyway, because the work does not arrive with the hours written on it.
There is a second problem. Prompts are almost always a lightly disguised version of a real problem the team is currently stuck on. Which means you are asking people to do unpaid work on your roadmap as the price of consideration. Most companies would deny this and most prompts fail the test.
What we did instead
Two changes, and they are cheaper than the thing they replaced.
A collaborative whiteboard session in the onsite. A real problem, sixty minutes, two people from the team in the room working it with the candidate rather than watching them. You learn things an artefact cannot show you. What they ask before they draw. Whether they change their mind when given new information. How they handle someone senior disagreeing with them. Whether they can put a bad idea on the board fast to get to a better one.
A portfolio review with the whole design team present. Three or four relevant projects, presented by the candidate, with everyone attending while the team was still small enough for that to be possible. We weren't after broad consensus. We were after exposure. The candidate meets the team as it actually is, and the team owns the hire rather than receiving it.
Both stages take the company's time rather than the candidate's, which I think is the correct direction for that cost to run.
The version I actually prefer
Everything above is what you do when you are hiring at volume and need a process a dozen people can run consistently. When I am hiring one person and have any say over how, I would rather do something else entirely. Work with them for a few weeks and pay them their rate.
Not a trial task. Not a scoped exercise marked against a rubric. A real piece of the actual work, with the actual team, at their normal hourly or daily rate, for two or three weeks.
You learn more in the first week of that than a full loop produces. How they ask for what they need. What they do on the day the brief changes. Whether the engineers start routing questions to them without being told to. None of that is available from a portfolio, and all of it is the job.
Objections are real, and I have made most of them myself. It is slow. It does not scale past a handful of hires a year. It needs a budget line and a contract that most companies have not set up. And it is hard for anyone still employed somewhere else. Run it as your only route and you select for people between jobs without anyone deciding to. That's its own bias, and a worse one than the take-home has.
So it is not a general answer. But where it is available I do think it beats everything else. It is the only format I know where the candidate is sizing you up on as good a set of facts as you have about them. Paying them properly is what makes that mutual rather than extractive.
What a portfolio stopped telling you
There's another reason to lean less on the work sample. Work samples have been telling you less every year for a decade.
Best portfolios are often not fully online any more. Work sits behind an NDA, or behind a password, or nowhere at all (individuals and companies do steal it, which is the unglamorous reason nobody says out loud). Senior people in particular often have a thin public presence and an out-of-date Dribbble. Treat that as a signal and you will routinely screen out the most experienced candidates in your funnel.
So the skill shifts. You are reading a career rather than a gallery: what they were hired to do, what shape the team was, who else was in the room, what happened after they left. Which brings its own failure, and I have written about that separately: the profile you teach a team to look for is partly a record of who was spared friction. Some of that reading comes from the résumé. Most comes from the conversation. The rest comes from reference calls, which I have argued are the most informative half hour in the whole process.
Both of the stages that replaced it spend the company's time instead of the candidate's, which is the right direction for that cost to run.
The cost of what I am recommending
I would rather state this than have it pointed out.
A live whiteboard session favours the fluent. It rewards people who think out loud, who don't mind being watched, and who do well with someone pushing back at them. Some outstanding designers are none of those things, and a take-home is the format in which they are at their best, because it lets them be slow and careful in private.
Those fixes are real but partial, and I'm not certain they close the gap. Send the problem in advance so nobody is doing cold recall. Keep the room to two people, never a panel. Say at the start that silence is allowed and then actually allow it, which requires the interviewers to tolerate discomfort they usually resolve by talking. Score the reasoning rather than the artefact on the board, and write the score before comparing notes with the other interviewer.
Even with all of that, you are trading one bias for another. My argument is narrower. Bias from a live session is smaller, more visible and more fixable than bias from a stage that ranks people by how much unpaid time they had that fortnight.
If you keep the assignment
Some teams will keep it, and there is one case where I would too. Hiring juniors and career-changers, where there is often no body of prior work to read and no reference who has seen the person operate. A short, paid, tightly scoped assignment may genuinely be the fairest thing available there.
If you keep it more widely than that, the conditions are strict. Pay for it at a real rate. Cap it at three hours and mean it, which means designing a prompt that is finishable in three. Make it a problem you have already solved, so it cannot be mistaken for free labor. Give at least a week of calendar to do the three hours in. And tell candidates what you are assessing before they start, because a rubric they cannot see is a test of whether they can guess your preferences.
Most companies do none of these and describe the assignment as the fairest part of their process (I have heard it called exactly that in a dozen debriefs).
I've built design and product organizations from first hire to dozens later, at Honor, Lyft, Grammarly and Pitch.
© 2026 Renato Valdés-Olmos