The Take-Home Test Is Dead. It’s Now a $20/Month AI Subscription Test, and You’re Being Graded Against Cheaters.

In June 2025, roughly 15% of candidates were using AI to cheat their way through assessments. By December, that number had jumped to 35%. That’s Fabric’s read on more than 50,000 candidates, and it means the take-home test you’re sweating over is now being graded against people who aren’t actually doing it.

Here’s our take: the take-home coding test, the async case study, the “just complete this at home” assignment, all of it is functionally dead as a fair signal. It didn’t die because companies stopped using it. It died because a $20 monthly subscription quietly turned it into an open-book exam that only the honest people are still taking closed-book. And if you think this only hurts cheaters, you haven’t seen how the detection dragnet catches everyone.

What Actually Broke

The take-home test was always built on one assumption: that the work you submit is your work, produced at roughly your real skill level. That assumption is gone.

Fabric’s team, which builds AI interview software, put it bluntly in their writeup on how AI cheating killed the take-home assignment. When a tool can produce a clean, working submission in minutes, the format stops measuring you and starts measuring your subscription.

• The signal collapsed quietly. One startup hiring manager made her take-home explicitly open-book with AI allowed, and reported the quality of submissions barely changed. Her theory: most people were already using it.
• Technical roles got hit hardest. In Fabric’s 19,368-interview study, 38.5% of all candidates showed signs of AI cheating, but in purely technical roles that number climbed to 48%. In sales, it was 12%.
• Juniors cheat at nearly double the senior rate. Candidates with 0 to 5 years of experience, the people most desperate to break in, are the ones leaning hardest on the tools.

Interview Guys Take: The uncomfortable part isn’t that cheating exists. It’s that the format failed silently. A take-home that everyone secretly aces with AI looks exactly like a take-home that’s working. Employers kept sending them out for months while the ground had already given way underneath.

Why This Punishes You Even If You Play It Straight

This is the part that should make you angry. You’re not competing against the honest applicant pool anymore. You’re competing against an inflated one.

When 61% of flagged cheaters in Fabric’s State of AI Interview Cheating dataset still scored above the pass line, that’s not a rounding error. That’s the majority of cheaters clearing the bar and pushing the bar higher for everyone behind them.

• The curve moves against you. If a third of the pool is turning in AI-perfect work, your honest, human-quality submission looks weaker by comparison, even when it’s a truer picture of your ability.
• Detection is a blunt instrument. Only 19% of hiring managers are extremely confident they’d catch a fraud, per the survey covered by Newsweek. A tool that misses most cheaters will absolutely produce false positives on real candidates.
• Suspicion becomes the default. When 59% of managers already assume someone’s faking it, you walk in guilty until proven human. That’s a brutal starting position for anyone whose real work happens to look polished.

The Tools Are Cheaper Than Your Coffee Budget

Let’s be specific about what’s on the other side of the screen. Products like Cluely, Interview Coder, and Leetcode Wizard run somewhere between $20 and $50 a month.

They use GPU-level overlays, DirectX on Windows and Metal on macOS, to render AI-generated answers directly on the candidate’s screen in a layer that screen sharing can’t capture. From the interviewer’s side, the candidate just appears to be thinking. It’s a teleprompter nobody else can see.

• This is a real market, not a fringe hack. The Cluely data breach in mid-2025 exposed 83,000 users, including records of which interviews each person used the tool in. That’s the scale we’re talking about.
• The risk isn’t only ethical. Those breached records are now a permanent reputational and legal liability for the people who used them. A $20 shortcut can follow you around.

Interview Guys Take: Anyone telling you to “just use the tools too” is skipping past the fine print. The vendors selling invisibility have already proven they can’t keep your data private. You’re trusting a company built on helping you deceive an employer to safeguard the evidence that you did.

Employers Aren’t Blind, They’re Overwhelmed

The people hiring you know exactly what’s happening. They just can’t keep up with it.

A Karat survey of 400 engineering leaders across the U.S., India, and China found 71% say AI makes it harder to assess a candidate’s real technical skills, a jump from the roughly 20 to 30% who’d have said that a few years ago. That data runs through the IEEE-USA breakdown of how AI is reshaping technical interviews.

• They still use take-homes anyway. Around 45% of U.S. employers keep using them even as confidence erodes, because nobody’s agreed on a replacement yet.
• The in-person pendulum is swinging back. Reporting puts the share of companies requiring in-person interviews to catch AI-assisted candidates at 72%, up from 24% in 2022. Even Google’s CEO has publicly pushed for a return to meeting people in a room.
• Video isn’t the safe middle ground it used to be. The same overlay tech that beats screen sharing works on camera calls too, which is worth remembering given how many companies now lean on video interviews.

The Defense Round Is The New Test

If there’s a survivor here, it’s not the take-home. It’s the take-home plus a live conversation where you defend it.

CoderPad’s State of Tech Hiring data, cited across recruiting coverage, shows 68% of companies now use take-homes and 41% run a hybrid: you submit the work, then sit through a 30 to 45 minute session explaining every decision. The logic is simple. If a tool wrote your code and you don’t understand it, the whole thing falls apart in five minutes of questions.

• Comprehension is the real signal now. Not “did you produce this,” but “can you explain why line 40 exists and what breaks if you remove it.”
• This rewards genuine preparation. Structure your walkthrough like a SOAR story: the Situation you were solving, the Obstacle in the requirements, the Action you took, and the Result. If you actually did the work, that narrative is effortless.
• It also raises the stakes on knowing your own resume. The same scrutiny is spreading to behavioral rounds, which is why understanding how few interviews each application actually yields matters. You can’t afford to fumble the ones you get.

Interview Guys Take: The defense round is quietly the fairest thing to happen to hiring in a while. It can’t be faked at scale, it rewards the people who did the work, and it turns “prove you’re not a cheater” into “show me how you think.” If you’re honest, you should want more of this, not less.

The Contrarian Read: Maybe The Job Changed, Not The Candidate

Here’s the argument that complicates the whole panic. Karat CEO Mohit Bhende pushes back hard on the idea that this is fundamentally about cheating.

His framing: the engineering job itself has changed, and the interview hasn’t. When nearly every developer uses AI tools daily and a large share of production code is now AI-generated, testing someone on their ability to work without the tool is testing a job that no longer exists.

• Meta already flipped the question. In late 2025 it piloted a coding round with GPT-4o mini, Claude, and Gemini available to candidates on purpose, shifting from “can you code without AI” to “can you direct AI well.”
• Some experts see cheaters as canaries. In a talent survey where 65% of managers were concerned about AI on assessments, one expert argued that AI use can flag a resourceful, high-potential candidate rather than a fraud.
• China’s already moved on. Chinese firms are nearly twice as likely as U.S. companies to allow AI in live technical interviews, having largely abandoned automated code tests for in-person sessions that watch how you reason.

What This Means Beyond The Tech Bubble

Don’t file this under “software engineers only.” The take-home test is a hiring tool across writing, marketing, analysis, design, and plenty of remote roles. The moment any of them can be done by a $20 subscription, the same collapse follows.

The practical fallout lands on real people making real moves. If you’re a career changer taking a pay cut to break into a new field, or a parent returning to work after time at home, you’re the person most likely to be graded against an inflated, AI-padded pool and least able to absorb an unfair rejection.

• In-person and roles without take-homes gain appeal. Hands-on interviews at places like a Home Depot or a nursing home judge you on presence, not on an unwatched submission.
• Experience is quietly becoming a moat. Seniors cheat at roughly half the junior rate, partly because they don’t need to. It’s one more reason the average new hire is now 42.
• The async work sample isn’t fully dead. Structured take-homes still appear to carry predictive value when paired with a defense. The unsupervised, honor-system version is the part that’s finished.

So where does this leave you. The unsupervised take-home is a broken measurement, and pretending otherwise helps nobody. If a company still hands you one with no live follow-up, read that as a signal about how carefully they’re actually evaluating people.

Focus your energy where the signal still lives: the live defense, the in-person round, the conversation where you explain your thinking out loud. That’s the ground a $20 overlay can’t hold, and it happens to be the ground where honest, prepared humans win. Get sharp at explaining your own work, because the resume that got you in the door still faces the same six-second scan it always has, and the interview on the other side is the part that’s changing fastest.