I spent a full weekend on a take home assignment for a Series B company. Built an end-to-end pipeline: ingestion, transformation, orchestration, tests, documentation. The instructions said "3 to 4 hours." It took 14. I submitted Sunday night. Monday, nothing. Tuesday, nothing. 2 weeks later, I followed up. Got ghosted. Never heard from them again.
That was 3 years ago. The problem has gotten dramatically worse.
53% of job seekers were ghosted by employers in 2026, up from 38% in 2024. 9% of those were ghosted specifically after completing a take home project or assessment. These aren't people who filled out an application and moved on. These are candidates who invested real hours, real thought, real labor into proving themselves, then received silence.
The data engineering interview loop has always been a gauntlet. DS&A, system design, SQL deep dives, behavioral rounds. But at least those happened in real time. You showed up, you performed, you got feedback (sometimes). The take home was supposed to be the humane alternative. Give candidates time. Let them work in their own environment. No whiteboard anxiety.
Instead, we got something worse.
The Whiteboard Died and Nothing Got Better
The anti-whiteboard backlash was legitimate. Solving a binary tree problem on a dry-erase board while 3 strangers stare at you is a terrible way to evaluate whether someone can debug a pipeline that silently drops 2M rows. Everyone agreed. The industry needed a better signal.
Take home projects were the answer. And for about 5 minutes, they made sense.
Then scope creep happened. A "2 to 3 hour" assignment became 8. Then 12. Then "build a full data pipeline with orchestration, testing, CI/CD, and a README that reads like production documentation." Companies started treating the take home not as a screen but as a proof-of-concept sprint. 45% of U.S. companies now use take home projects in their hiring process. 47% of hiring managers prefer them over live coding for mid-level roles.
The problem isn't the format. A well-scoped 90-minute take home with a code review follow-up produces some of the highest signal of any interview format. The problem is that "well-scoped" has become the exception. Time estimates are systematically dishonest: tasks marketed as "a few hours" routinely consume 12 to 20. One candidate documented canceling a weekend trip to finish an assignment, only to be ghosted for 5 weeks afterward.
And here's the kicker: 71% of engineering leaders now say AI has made technical assessment "meaningfully harder," with take homes suffering the worst signal degradation. 62% of candidates already use AI tools during interviews regardless of the rules. So the format that was supposed to replace the arbitrary whiteboard is now compromised by the same forces that made whiteboard performance meaningless.
The industry didn't solve the interview problem. It transferred the cost from companies to candidates, called it progress, and moved on.
Who Gets Filtered Out of Data Engineering Hiring
A 12-hour take home selects for people with 12 free hours. That's it. It doesn't select for the best data engineers. It selects against senior engineers with families, against people interviewing at multiple companies, against anyone who values their time proportional to their experience.
The numbers back this up. 40 to 60% of senior engineer candidates drop out of take home assessment processes. At Dropbox, before they pivoted away from the format, 20% of candidates simply never submitted. The primary withdrawal reason from senior candidates? "Too much unpaid time" and "another company finished my loop in 1 week."
This is the opposite of what hiring should do. You're not filtering for quality; you're filtering for availability. The engineer with 10 YOE, 2 kids, and a demanding job isn't going to spend her weekend building your toy ETL pipeline for free. She's going to take the offer from the company that respected her time with a 90-minute live session.
80% of surveyed engineers believe take homes should take 4 hours or less. 58% believe they deserve payment. Only 4% have ever received it.
Here's what makes it worse for data engineering specifically: the assignments aren't generic. A frontend take home might be "build a todo app." A data engineering take home is "ingest this CSV, model it into a star schema, orchestrate daily refreshes, handle late-arriving data, write tests, document assumptions." That's not a screen. That's a sprint.
The "unlimited time to show your best work" framing is a trap. High-conscientiousness candidates, the ones you actually want to hire, over-invest in polish, testing, and documentation. A "3-hour task" becomes a weekend because they can't submit something they'd be embarrassed by. The format punishes exactly the trait you're screening for.
Ghosted After 20 Hours: the Feedback Black Hole
94% of candidates want feedback after an interview. 5.5% receive it.
You can frame take home assignments as evaluation tools. You can frame them as equitable alternatives to whiteboard hazing. But when you hand someone a 20-hour project, they complete it, you reject them, and you give zero feedback, you've extracted labor for nothing. The candidate calls it getting ghosted. Some are starting to call it spec work.
The asymmetry is staggering. An employer can ask 10 candidates to do a take home project in a week: 70 hours of candidate labor total, 1 to 2 hours of employer review. And the worst part? 79% of candidates would reapply to a company that rejected them if they received constructive feedback. The feedback gap isn't just rude; it's bad strategy. Companies are burning their own candidate pipeline because writing "we went with someone whose modeling approach aligned more closely with our stack" takes 45 seconds and nobody will do it.
Interviews per hire jumped 42% since 2021, from 14 to 20 per position. Time to hire increased 24%. The process is getting longer, the feedback is getting sparser, and the labor ask is getting bigger. 72% of job seekers report negative mental health impacts from long hiring processes and poor employer communication. That's not a statistic; that's a gut punch.
And then there's the legal question that nobody wants to ask out loud. Under the Fair Labor Standards Act, anyone performing real work that benefits an employer must be paid at least minimum wage, even during a trial. The DOL successfully recovered $50K in back wages from a company that disguised unpaid candidate "working interviews" as applications. Most take home ghosting stems from hiring process chaos (role closure, budget freeze, manager turnover) rather than deliberate code theft. But the fact that the question even comes up tells you everything about how broken trust has become.
Without a signed agreement, you retain copyright on your take home submission. "Retain copyright" and "can prove a company shipped my transformation logic" are very different things, though. The spec work accusation may be overstated for most roles. But when a company asks you to build something on their proprietary dataset, using their business rules, solving a problem that looks suspiciously like a feature on their roadmap? That's not evaluation. That's consulting. And consulting gets invoiced.
What Good Actually Looks Like (and How to Protect Yourself)
Stripe's SQL Bug Squash interview is 45 to 60 minutes. The candidate debugs 4 to 5 broken production SQL queries live, collaboratively, with an interviewer. It isolates debugging skill, which is what data engineering actually is most of the time. It sidesteps the "was this AI-generated?" problem entirely because you're watching someone think in real time. And it takes an hour, not a weekend.
The ethical benchmark for take homes already exists: 3 to 4 hours of scoped work, delivered within a 48 to 72 hour window, with a committed feedback timeline. This isn't novel. It's the standard most companies acknowledge and then violate.
Live pair-programming with AI tools allowed is emerging as the strongest replacement: 60 to 90 minutes, interviewers observe real-time workflow and decision-making. You can't fake your way through a live debugging session with Copilot; the interviewer sees how you prompt, how you evaluate suggestions, how you reason under pressure. That's signal. A polished take home submission tells you someone (or something) can write clean code. It tells you nothing about how they work.
If you're preparing for these kinds of loops and want to sharpen the fundamentals that actually get tested, we built the prep around exactly this reality; check DataDriven for etl interview questions that mirror real debugging and modeling scenarios, not toy problems that waste your time twice.
But prep is only half the equation. You also need to protect yourself before you start.
Ask for the feedback commitment upfront. Before you accept the assignment, ask directly: "Will I receive written feedback regardless of the outcome?" If they hedge, that tells you everything. A company that won't commit to 5 minutes of feedback after asking for 10 hours of your time has already shown you how they value the exchange.
Sequence matters. Don't do the take home before initial interviews. Talk to the team first. Get a read on culture, on the role, on whether you'd even want to work there. If you invest 15 hours and then discover in the onsite that the "data engineering" role is actually an analyst position with a pipeline on the side, you've wasted a weekend.
Time-box ruthlessly. If they say 4 hours, spend 4 hours. Submit what you have. Add a README section called "What I'd do with more time." If they reject you for not gold-plating beyond their own stated scope, that's a company that will expect 60-hour weeks and call it "ownership."
Negotiate the format. About half of candidates dislike take homes enough to drop out. You can propose alternatives: a portfolio walkthrough of production work you've already built, a shorter time-boxed replacement, a live pairing session. Some companies treat refusal as disqualifying. Those are the same companies that will ghost you after you submit. You're not losing much.
The take home interview isn't inherently broken. A tight, well-scoped, 3-hour assignment with a feedback guarantee and a code review follow-up is genuinely good signal. But that's not what most companies are running. What most companies are running is a 20-hour unpaid work trial with no feedback, no respect for your time, and a 53% chance of total silence.
The format was supposed to be the humane alternative. Right now, it's just the whiteboard in a nicer suit.
What's the worst take home you've been asked to complete, and did you ever hear back?
Top comments (0)