I got sent a take-home assignment last year that asked me to build an end-to-end pipeline: ingest from three APIs, transform in Python, load to a warehouse, write tests, document my design decisions, and prepare a 15-minute presentation for "the team." The recruiter said it should take "about four hours." I timed myself. It took fourteen. I didn't get the job. I didn't get feedback. I got a form rejection three weeks later.
That's not an interview. That's a consulting engagement with a 0% billing rate.
The Scope Creep Nobody Talks About
Take-homes weren't always like this. The original pitch was reasonable: instead of whiteboard hazing where you reverse a linked list while someone watches you sweat, you get to work in your own environment, at your own pace, on something resembling real work. That was the deal. A couple hours, a focused problem, maybe a short discussion afterward.
Then companies got greedy.
The recommended best practice is still 2 to 4 hours. Over 80% of survey respondents believe take-homes should cap at four hours. But candidates consistently report spending 5x the stated estimate. Companies will write "don't spend more than 3 hours on this" at the top of a prompt that includes building a working MVP, writing a README with architecture docs, recording a demo, adding unit and integration tests, and documenting your trade-offs. That's not a 3-hour task. That's a small freelance project.
The scope expectations in 2026 are indistinguishable from paid contract work. One candidate reported being asked to create a 30-minute learning module with video, graphics, voiceover, and interactive elements. The estimated freelance market value? $2,800. For an "assessment."
Here's what's actually happening: data engineering take-homes have quietly evolved from "show us you can write SQL" to "build us a proof of concept we might actually use." And the line between those two things is the line between an interview and unpaid labor.
58% of engineers believe they deserve payment for take-homes. Only 4% receive it. Read those numbers again.
If a candidate invests 15 hours with a 10% chance of advancing, the expected return per hour is zero. That's not an interview process; that's a lottery where you pay with your weekend.
Free Consulting in Disguise
Let's talk about the part nobody wants to say out loud: some companies are using candidate submissions.
Indeed's own hiring research flags the concern directly: "Companies may even steal the ideas of candidates, use them, and not give credit or compensate the candidate." That's not a fringe take from a disgruntled Reddit poster. That's on a major job platform's hiring guide.
The structural problem is simple. When you ask a data engineer to build a pipeline that ingests your actual data format, transforms it according to your actual business logic, and loads it into your actual warehouse schema, you've crossed the line from evaluation to extraction. The candidate doesn't know if their code will ship or be discarded. The company doesn't disclose what happens to submissions. The information asymmetry is total.
And it gets worse. About 50% of job seekers strongly dislike take-homes and drop out entirely. But here's the paradox: Dropbox found that 20% of candidates abandoned their process before completing assignments, and the ones who dropped out were often the strongest candidates. They had competing offers. They had leverage. They didn't need to grind 15 hours for a maybe.
So who finishes? Candidates without alternatives. The desperate. The junior engineers without competing offers. The people who can't afford to say no. The take-home isn't filtering for talent; it's filtering for availability.
This is especially brutal for marginalized candidates. If you're working a second job, handling caregiving, or don't have reliable internet access, a 15-hour unpaid assignment isn't a minor inconvenience. It's a gate that has nothing to do with whether you can do the work. The equity fracture is real: unpaid take-homes create a class-based filter where financial stability determines who gets to compete.
76% of recruiters say take-homes improve hiring quality. Nobody surveyed the candidates to ask if they felt fairly assessed. Funny how that works.
Red Flags Before You Even Open the Repo
After doing somewhere around 20 interview loops in a single job search (some went well, some went laughably poorly), I've developed a pretty reliable radar for take-homes that are going to waste my time. Here's what I look for now.
No time estimate at all. If the prompt doesn't tell you how long they expect it to take, they either don't know or don't care. Both are bad. A company that can't scope a 3-hour exercise is telling you something about how they scope projects internally.
The deliverables list is longer than the problem statement. When the requirements section says "build a pipeline" but the deliverables section says "working code, tests, documentation, architecture diagram, trade-off analysis, recorded demo, and a 15-minute presentation," you're not being evaluated. You're being outsourced.
The data looks suspiciously like their actual data. Generic datasets (public APIs, sample CSVs) are fine. When the schema matches their product domain a little too closely, when the transformations feel like real business logic, that's not a coincidence. That's a proof of concept with plausible deniability.
No compensation and no timeline. The gold standard for a fair take-home is 90 minutes of focused work plus a 30-minute walkthrough. If they're asking for more than 4 hours and offering nothing in return, the math doesn't work in your favor. Labor experts agree: unpaid assignments exceeding 2 to 3 hours cross into territory where compensation is ethically and legally justified.
They ghost after submission. 8 in 10 hiring managers admit to ghosting candidates. If you're investing a full weekend into something, you deserve feedback. A company that can't write two paragraphs about why they passed on you after you wrote two thousand lines for them is telling you exactly how they'll treat you as an employee.
The AI angle makes all of this worse, not better. Cheating rates on take-homes jumped from 15% to 35% in six months. Companies are responding not by making the process fairer, but by making it harder and longer. Classic arms race. The format's legitimacy is collapsing in real time.
How to Push Back Without Burning the Bridge
Here's what I've learned the hard way about career management in the interview game: asking for scope clarity isn't weakness. It's the move that separates you from candidates who'll silently overcommit and resent it.
Before you open your IDE, send this: "I want to make sure I'm aligned with your expectations. I estimate this will take X hours based on the requirements. Is that consistent with what you've seen from other candidates?" That's it. Professional, direct, and it forces them to put a number on the record.
If the number they come back with is wildly different from your estimate, you have information. Either the scope is genuinely smaller than it looks (great, clarify what's optional) or they're lowballing the time estimate to avoid scaring you off (red flag).
If the assignment exceeds 4 hours, it's completely reasonable to ask about compensation. Buffer runs 45-day paid trial projects. Webflow uses 3 to 5 day paid contract follow-ups. These aren't charity; a paid engagement costs the company less than a bad hire. If a company balks at paying for 8 hours of your time, they're telling you the ROI math doesn't work, which means they're sending this to dozens of candidates and hoping one sticks.
And here's the contrarian take that took me years to internalize: if a company demands 20 hours for an initial screen and you walk away, you didn't lose an offer. You dodged a scope-creep culture. The real risk isn't declining; it's spending 20 hours and still losing to someone who had better chemistry in the 30-minute walkthrough. That's not career management. That's gambling with your time as the chips.
What Fair Actually Looks Like
Fair take-homes exist. They're just rare.
The format works when companies respect it: 90 minutes of focused work, a clear rubric, a 30-minute walkthrough to verify you wrote it and can explain it. Well-designed take-homes show 35% higher correlation with job performance compared to whiteboard interviews. The signal is real. The abuse is the problem, not the concept.
Companies that compensate candidates and set clear requirements see 85%+ completion rates. The ones that don't? 60 to 70%. The data is screaming the answer. Pay people for their time, scope the work honestly, and the format is better than every alternative.
900+ companies are listed on the Hiring Without Whiteboards repository, explicitly committing to fair evaluation. Some use pair programming on real codebases. Some use portfolio reviews. Some use short, focused take-homes with hard time caps. The alternatives exist. The industry just hasn't decided to care yet.
The uncomfortable truth is that when hiring fundamentals are correct (clear rubrics, structured evaluation, diverse interviewers), the format almost doesn't matter. Take-home, live coding, pair programming; they all work when the process respects the candidate. They all fail when it doesn't.
I've been on both sides of the interview table. I've been the candidate grinding through a 14-hour take-home for a form rejection. I've been on hiring panels where we evaluated submissions in under 10 minutes that took candidates an entire weekend. Both of those experiences made me angry for the same reason: the asymmetry is the point, not the bug.
So here's my question for anyone who's been through this recently: what's the most egregious take-home you've been asked to do, and did you finish it or walk away?
Top comments (0)