Tuesday, November 3, 2026

Why candidates abandon your take-home, and the five objections to fixing it

Josh Priollaud

Your best candidates are dropping out at the take-home step. It isn't because they're lazy, and it isn't only because the assignment is too long.

It's because you're asking for free work with no terms behind it. The candidate doesn't know whether they'll be paid, whether they can show the work to anyone else, or whether you'll quietly use it. The most skilled candidates have other opportunities, and they do the math and move on. The ones who stay are the hungriest, the people who'll do anything for a shot. Hunger is a fine quality, but it isn't the same as skill.

The fix isn't a shorter assignment. It's a neutral platform with fair terms: a promise that the hiring company won't use the candidate's work commercially, portfolio rights for the candidate, an honorarium that pays out automatically, and a short agent-led video walkthrough where the candidate explains their own work. That's what Pudding does.

When I describe this to hiring teams, I hear the same five objections. Here they are, with the reasoning behind my answers.

"We'll build this in house. We're AI pros."

Go for it. Here's the build list.

A live video agent. Not a chatbot. It needs to hold a spoken conversation in real time, follow the candidate's screenshare, read the code, doc or design they're showing, and ask a sensible next question about what's on screen.
Artifact ingestion. The agent has to know the submission before the call starts: a repo, a Figma file, a memo, a spreadsheet. Every format is its own parsing problem.
Payments, end to end. Paying a candidate $200 sounds simple until it's a vendor-onboarding form, a tax form, a currency conversion and net-45 terms. If payment takes six weeks and an invoice, it stops reading as respect.
A two-sided product. The hiring team needs briefs, review and recordings. The candidate needs a clean way to submit, record and get paid. Both sides have to be polished, because candidates judge you by the tool you send them.

All of that is buildable. The part you can't build is trust.

Even with a perfect in-house tool, the terms are still yours. You're asking candidates to trust the same company that's evaluating them not to use their best ideas. Your legal team writes the terms, and your legal team works for you. A neutral third party with standard terms is a different promise, the same way escrow is a different promise from "just wire it to us, we're good for it."

"We're never going to pay candidates for take-homes."

Then you'll keep losing the top of your funnel.

Look at it from the candidate's side. An unpaid take-home is a few hours of real work with no pay, no terms and no guarantee the role is even still open. A skilled candidate with three options compares that ask with the companies that didn't make it, and drops the one that asked for free work.

The numbers bear this out. Indeed's Interview Tasks 2026 survey of employers and job seekers, as reported by HCA Mag and Dynamic Business, found:

73% of employers said they'd lost candidates after setting an interview task.
66% of job seekers said they'd withdrawn from an application, or decided not to apply at all, because of an interview task.
93% of job seekers said a task should be paid if it takes three hours or more. Only 36% had ever been paid in full for one.

Engineers report the same pattern. interviewing.io surveyed almost 700 of its users and estimated that candidates complete take-homes about 62% of the time. Its advice: if fewer than 85% of candidates complete yours, it's time to make a change, because "losing more than 15% of your candidates to attrition at any given stage in the funnel is bad."

Look at who finishes. In the same survey, 70% of people who completed a take-home said they did it because they "really wanted to work at the company and were willing to do what it took." That's hunger, and a famous company can rely on it to fill a shortlist. Most companies can't, and the candidates who weren't hungry enough to work for free include plenty who were good enough to hire.

You'll also hear about it in public. Search Reddit for "unpaid take-home assignment" and you'll find a steady stream of candidates describing long unpaid assignments, being ghosted afterward, and suspecting their work was used. It isn't only Reddit. One data engineer's post on DEV, "They Gave Me a 20-Hour Take-Home. I Did It. I Got Ghosted.", describes an assignment billed as "3 to 4 hours" that took 14, followed by silence. Posts like these often name the company. They rank in search, and the next candidate who Googles your interview process will find them.

Pudding uses AI to check each hiring team's assignment guidelines and rejects them if the time estimate is unrealistic.

Candidates have noticed the gap between what they think is fair and what they get. In the interviewing.io survey, 58% of engineers said they deserve compensation for completing take-homes, and only 4% had ever received it.

Paying fixes all of this at once. A few hundred dollars per finalist is small next to a recruiter's fee or a month with the seat empty. There's a longer version of this argument in Are unpaid take-home assignments exploitative?

"Candidates love our unpaid take-homes."

How would you know? Feedback from candidates whose identities aren't masked is feedback under duress. They may want to work for you someday, and they won't risk their reputation by telling the company that's evaluating them that its process is unfair.

Ask candidates somewhere they can't be identified and you hear something different. In Indeed's survey, 55% of workers said unpaid interview tasks are unfair, and 27% called them "very unfair." In the interviewing.io survey, many engineers who finished a take-home had such a poor experience that they said they'd never apply to certain companies again.

And the most unhappy candidates often never tell you anything. They're the ones who never submitted it, so they don't show up in your survey.

"Our candidates won't do an AI interview."

That worry is fair. Greenhouse's 2026 Candidate AI Interview Report found that 38% of US candidates have walked away from a hiring process because it included an AI interview, and another 12% say they would.

But look at what they're walking away from. The top triggers were "pre-recorded video interviews scored by AI with no human present" (33%), companies not disclosing how AI would be used (27%), and AI monitoring during the process (26%). Only 19% want less AI in hiring. What candidates ask for is guardrails: 38% want to know a human reviews the AI's evaluation before any decision is made.

The Pudding walkthrough is different on the points that matter, in three ways.

They get paid. The walkthrough is part of a paid assignment, not an unpaid gate in front of a human.

The agent doesn't judge quality. It doesn't score the work or decide who moves forward. People on your hiring team do that, with the work and the recording in front of them. The agent's job is to get the candidate talking about authorship and thinking, which is the part a finished file can't show. Questions are along the lines of:

"Walk me through what you changed and where you started."
"Why this approach rather than the obvious alternative?"
"Which part was hardest, and how did you get unstuck?"
"What did you use AI tools for, and what did you write or decide yourself?"
"What would you do next with another day?"

They keep the work. The terms give the candidate portfolio rights, so the assignment is something they can show the next company, not something that disappears into your drive.

I ran it on my own code from past jobs. It felt like explaining my work to a curious colleague, not like being graded.

This matters more now that candidates can use AI on any take-home. That's fine, because they'll use it on the job. What you need to know is whether they understand what they shipped, and a short conversation about the work answers that faster than anything else. There's more on why in How many words does each hiring step give you?

"We still won't pay. We want candidates who'll sacrifice just for a chance to work with us."

Then you're selecting for the hungriest candidates, not the most skilled.

The hungriest candidates will do your take-home for free. They're early in their careers, between jobs, or desperate for a break, and a chance to impress you is worth a weekend to them. There's nothing wrong with that. But the most skilled candidates are usually somewhere else: doing well in a job they don't need to leave, weighing other opportunities, and with enough self-respect to say no to free work. They're the ones an unpaid take-home filters out.

So a candidate's willingness to sacrifice tells you how badly they want a job. It doesn't tell you how good they'll be at it.

It also isn't a sacrifice that buys the candidate any protection. Look at the terms candidates accept on the way to your take-home. The platforms tell candidates they own their work, then take a broad license to it for themselves:

HackerRank's candidate terms give HackerRank a "perpetual, nonexclusive, royalty-free, fully paid up, sublicensable, transferable right and license" to submitted code "for commercial and non-commercial purposes," including permission "to permit third parties to use, your Content for the purpose of training" AI models. They add that submitting work does not "establish any confidential relationship with HackerRank."
Indeed's terms count resumes and messages as User Content and take a "perpetual, fully paid, royalty-free, transferable, and sublicensable" license to it, including to "develop, test, train, and improve machine learning and AI capabilities."
CodeSignal's terms have each candidate grant the hiring company "a non-exclusive, perpetual license to access such Candidate's User Submissions."
LinkedIn's User Agreement takes "a worldwide, transferable and sublicensable right to use, copy, modify, distribute" whatever members upload, "without any further consent, notice and/or compensation."

None of these terms say whether the hiring company can use the work itself, for example by shipping the code or running the campaign. That isn't their job. Applicant tracking systems say outright that the employer is in charge: Greenhouse says its customers "are the controllers" of candidate data and manage it "in accordance with its own privacy policies," and Workday says the same for its enterprise customers. So once a candidate hands you a marketing plan, a pricing recommendation or a fix for your checkout flow, the only limit on what you do with it is your own good intentions.

Candidates know how often that isn't enough. In the same Indeed survey, 45% of employers admitted reusing work a candidate had submitted, and 37% of workers believed it had happened to them. That's the terms-of-service gap.

Pudding's terms close it. When a candidate completes an assignment on Pudding, the hiring company agrees not to use the candidate's work commercially. You can't ship their code, run their campaign or adopt their pricing plan. The candidate, meanwhile, keeps the right to show the work in their portfolio. Without those two promises in writing, a strong candidate's rational answer to an unpaid take-home is no.

If you want to know who's good, pay them and look at what they do with three well-scoped hours. Skill shows up in the work, not in the willingness to work for nothing.

Try it on your own role

The fastest way to see the difference is to run one. Set up a sandbox in Pudding, write a brief for a role you're hiring for, and go through it as the candidate: submit the work, record the walkthrough, and see what your hiring team would get.

Set up a sandbox →

One more question to take back to your team: what's your take-home completion rate? If you don't know, that's the first problem.

Tuesday, November 3, 2026

Why candidates abandon your take-home, and the five objections to fixing it

"We'll build this in house. We're AI pros."

"We're never going to pay candidates for take-homes."

"Candidates love our unpaid take-homes."

"Our candidates won't do an AI interview."

"We still won't pay. We want candidates who'll sacrifice just for a chance to work with us."

Try it on your own role

Related reading

Product

Company

Support

Legal