I got a take-home assignment last year from a company I was genuinely excited about. "Should take about four hours," the recruiter said. Build an ingestion pipeline, model the data, write tests, document your design decisions, and prepare a 15-minute presentation walkthrough for the panel. Four hours. I laughed, closed my laptop, and started on it the next morning like it was a sprint. Sixteen hours later I had something I was proud of. Clean pipeline, solid tests, real documentation. I submitted it on a Sunday night. Monday I got a form rejection. No notes. No feedback. Not even which stage I failed. Just "we've decided to move forward with other candidates" and a link to their Glassdoor page.
That was the moment I stopped pretending take-homes are assessments. They're consulting gigs. Unpaid ones.
The Scope Creep Nobody Talks About
Five years ago, a data engineering take-home was a focused exercise. Model this dataset into a star schema. Write a few SQL transforms. Maybe a short README. Two to four hours, tops. Bounded, reasonable, and actually useful for evaluating how someone thinks about data.
That version is dead.
Today, 68% of companies use take-home tests, up 12% year over year. And the scope has quietly ballooned into something unrecognizable. Full pipeline implementations. Test suites with coverage thresholds. Documentation that reads like a design doc. A presentation follow-up where you defend your architecture to a panel. We're talking 10 to 20 hours of work, routinely, for a role you haven't been offered.
Industry best practice caps take-homes at 90 minutes of expected effort. The reality? Candidates consistently take 2x longer than company estimates to reach submission quality. That "four-hour" assignment is an eight-hour assignment. That "weekend project" is a week of evenings. And 25% of companies are still handing these out like they're reasonable asks.
Here's the part that makes my eye twitch: 71% of engineering leaders openly say take-homes no longer generate useful signal. AI has degraded the format so completely that leaders themselves rate take-home signal as "degrading fastest" among all assessment types. They know it's broken. They keep doing it anyway.
The attempted fix is even worse. Companies panicked about AI usage and responded by inflating scope. The logic, if you can call it that: make the assignment so large that AI can't do it alone. Except longer assessments don't defeat AI; they defeat candidates. Candidates with kids. Candidates working full-time jobs. Candidates from non-traditional backgrounds who can't burn 20 hours on a maybe. One candidate documented spending 32 hours on a single assignment, then got rejected for omitting a feature that was never mentioned in the requirements. Another was asked to build a learning module that would've billed at $2,800 as freelance work.
A four-hour take-home is a fair test. A 20-hour take-home is free consulting dressed up as an interview.
59% of job seekers now say unpaid take-home assignments are the number one reason they won't apply. Not comp, not culture, not location. The assessment itself is the dealbreaker.
AI Banned, Rubrics Unchanged
Two thirds of companies ban AI use in their interview process. Sounds decisive. Except fewer than 30% of those companies have actually updated their assessments or retrained their interviewers. They slapped a "no AI" sticker on a 2015-era take-home and called it policy.
The enforcement gap is almost comical. One company measured 80% of candidates using LLMs on take-home tests despite an explicit prohibition. AI cheating on take-homes doubled from 15% to 35% between June and December 2025. In purely technical roles, 48% of candidates show signs of unauthorized AI use. The ban is a suggestion, not a guardrail.
Meanwhile, the rubrics these companies grade against were built to evaluate raw coding speed and syntax accuracy. Those signals collapsed the moment Claude could produce a clean solution in seconds. But nobody rewrote the rubric. Nobody redefined what "good" looks like when the baseline output quality shifted. Hiring managers score problem-solving and architecture judgment, but the assessment they hand out measures code-from-scratch, a skill that's now commodity.
The split in the industry tells you everything. Meta and Shopify openly invite AI tools into their assessments. They've decided to test "can you use AI well" rather than "can you code without it." Goldman Sachs and Amazon maintain hard bans for candidates while investing heavily in internal AI tools for their own engineers. The hypocrisy is so blatant it's almost impressive. You can't use AI to get hired here, but once you're in, you'd better use it or you're slow.
Banning AI in interviews creates a discontinuity between evaluation and production. In 2026, writing code without AI assistance is the exception, not the norm. You're testing candidates in an environment that doesn't reflect the environment they'll work in. That's not assessment; that's theater.
70% of You Will Never Hear Why
Here's the stat that should make every hiring manager uncomfortable: 69.7% of candidates receive zero feedback after rejection. Not "insufficient feedback." Zero. Nothing. A form email and silence.
61% of candidates report being ghosted entirely after interviews. No rejection, no closure. Just silence from a company that asked them to spend a weekend building a pipeline.
Companies hide behind legal risk. "We can't give feedback because candidates might sue." This is, to put it plainly, nonsense. Employment law distinguishes between subjective rejection reasons ("you seemed low-energy") and factual, role-specific feedback ("your schema migration approach didn't handle the edge case we were testing for"). The second type is almost litigation-proof. No engineer has successfully sued a company over constructive technical feedback. The legal defense is a myth that compliance teams perpetuate because "say nothing" is the lowest-variance strategy. It's organizational laziness wearing a legal costume.
The business case against silence is overwhelming. 79% of candidates would reapply to a company if they'd received feedback. Recruiters who share feedback see a 126% increase in candidate referrals. Companies withholding feedback aren't just being rude; they're burning bridges they'll need to cross again in 18 months when they're hiring for the same role.
But here's the real cruelty. When the assessment demands 10 to 20 hours, and the rejection carries zero feedback, you've extracted labor and returned nothing. Not compensation, not signal, not even a paragraph explaining what to work on. The candidate can't even reuse the learning because there is no learning. It's labor arbitrage dressed up as a career opportunity.
Only 17% of external candidates receive feedback, compared to 65% of internal candidates. If you already work there, you get a debrief. If you're on the outside spending your weekend on their assignment, you get a template. The double standard is institutional.
What Actually Works
The good news: some companies figured this out. The better news: it's not complicated.
Live debugging interviews, running 60 to 90 minutes, are replacing puzzles at companies like Cloudflare, Datadog, and GitHub. Candidates get a broken system. They debug it. Interviewers watch the process: how do you form a hypothesis, how do you narrow the search space, do you narrate your thinking. You're evaluated on engineering judgment, not memorization speed. A candidate who thinks aloud and corrects wrong hypotheses scores higher than one who guesses fast but can't explain why.
For senior and staff roles, pair programming on a debugging or refactoring task is the highest-signal round you can run. Forty-five minutes, real code, real collaboration. It surfaces the kind of judgment that 20-hour take-homes never could, because judgment shows up in conversation, not in a solo sprint nobody watches.
Uber runs a two-hour on-site schema critique instead of toy problems. Stripe bounds their take-homes to one to three hours with clear scope. Both companies report higher completion rates and better signal than the bloated formats they replaced.
The pattern is obvious: bounded time, realistic work, human interaction. If you want to know how someone debugs a broken DAG, hand them a broken DAG and watch. Don't ask them to build one from scratch over a weekend and then ghost them.
If you're a candidate stuck grinding through these loops, focus your prep on the concepts that transfer across every format: data modeling, pipeline architecture, query optimization. I've found that a resource like datadriven.io is good for etl interview questions if you want structured reps on the technical fundamentals without wading through another generic course. The game is arbitrary, but the concepts compound regardless of which format a company throws at you.
The System Knows It's Broken
72% of job seekers report negative mental health impacts from lengthy hiring processes. Candidate ghosting hit a three-year high in 2026. The market has 2.2 million fake openings monthly, candidates respond with AI-powered mass applications, companies respond by banning AI, and the entire system spirals further from producing any useful signal for anyone.
The profession acknowledges the assessment is unreliable while refusing to stop using it. This isn't a transitional phase. It's institutional paralysis. Companies would rather extract 20 hours of free work from someone they'll reject silently than spend 90 minutes on a live session that actually reveals how an engineer thinks.
I've been through enough of these loops to know the system doesn't reform itself. It changes when candidates refuse to participate and when hiring managers with enough authority say "this is stupid, let's stop." If you're in a position to design an interview process, bound the time, provide feedback, and evaluate how people think, not how much free labor they'll tolerate.
If you've done one of these 20-hour take-homes recently: what was the assignment, and did you hear anything back?
