What this stage is
A homework assignment sent between screening and the technical sections: build a small application, implement a service, or dig into someone else’s legacy code and fix it. It differs fundamentally from live coding and the algorithms section. There, code is written in front of the interviewer and the reasoning is what gets assessed. Here, code is written at home and alone, and what gets assessed is the submitted result, whose authorship cannot be verified.
Why the format lost trust
The collapse in trust came from the reviewers’ side. In February 2026 CoderPad surveyed around 650 people involved in hiring — developers, recruiters and hiring managers. Only 12% of recruiters said take-homes reflect a candidate’s real ability to do the job. Asked about cheating, 52% of them named solutions generated by AI agents as the main risk.
Candidates confirm the concern. In an Anketolog survey conducted in June 2026 among 1,800 people who had looked for work through online platforms, 30% had asked a neural network to help with a take-home. Another 28% had used one to prepare answers to standard interview questions.
Developers themselves are not writing the format off. In CoderPad’s 2025 survey, 46% named the take-home one of the best ways to demonstrate their skills — second only to a technical conversation. Opinions split along the two sides of the table: those being assessed find the format convenient, while those doing the assessing don’t consider it a reliable signal.
There has been no public abandonment of the format. No major company — neither in Russia nor among US big tech — has dropped take-homes citing AI. Large-scale Russian hiring of experienced developers was live even before neural networks. The only explicit statement came from the small company SSP SOFT in January 2025, which called the take-home “a vestige of the hiring process” on the grounds that “solutions to take-home assignments are no longer searched for — they are generated” (in Russian).
How developers feel about it
Programmers are the profession most likely to refuse, and by a wide margin. In February–March 2025 SuperJob surveyed 1,400 people, 200 per professional group. 19% of programmers always complete a take-home, 54% sometimes, and 27% never apply to vacancies that require one. Over four years the share of refusers grew by 7 points. For comparison, 63% of QA engineers answer “always,” as do 46% of designers and 30% of marketers.
This survey gave rise to a figure that gets misquoted. “Only two programmers in ten apply to vacancies with a take-home” is a distortion. 19% are those who always do the take-home. Another 54% do it sometimes, and together they make up almost three quarters who agree at least occasionally. “Two in ten” is what you get by taking only “always” and discarding “sometimes.”
How many hours is reasonable
Both sides converge on two to four hours, and they arrive there independently. In an interviewing.io survey of around 700 engineers, more than 80% named four hours as the ceiling, and the most common answer was two. A career consultant at hh.ru sets a similar benchmark: a take-home should not take longer than one interview, that is, an hour and a half to two hours, and anything over four hours is already a reason to discuss payment or a smaller scope.
An AGIMA survey (214 people, October 2023) gives similar thresholds, but it needs to be read carefully: 90% of respondents called an unpaid assignment of more than eight hours unacceptable, and 53% said the same of more than four. That is an attitude, not behavior: in the same survey 89% said they were willing in principle to complete an unpaid take-home. The sample was mixed — developers, managers and analysts — so these figures cannot be applied to developers alone.
Where a test becomes work
Russian law does not require take-homes to be paid, so everyone draws the line between a skills assessment and unpaid labor for themselves. A career consultant at hh.ru reduces it to one question: “Could the company take your result tomorrow and use it with minimal changes? If so, that’s work, not an assessment” (in Russian). By this logic, what should be paid for is anything with commercial value: working code that will go into the product, a mockup that will ship in a release, copy that will be sent to a client.
The gap between expectation and reality has been measured: in the interviewing.io survey, 58% of engineers believe take-homes should be paid, and 4% had been paid for one. Russian data on the share of companies that pay for take-homes is not publicly available at all.
The signs of an assignment that has crossed the line repeat from company to company: it is sent before any conversation with the team, it reproduces a specific piece of the product, the scope amounts to several days of work, and the assessment criteria are not stated.
How take-homes are actually reviewed
The belief that nobody reads take-homes is widespread but not supported by data. It rests on individual opinions — a senior frontend developer at Doubletapp puts it bluntly: “few people review them properly, and most assignments are never even opened” (in Russian). No one has taken measurements that would confirm or refute it. A counterexample is on record: SKB Kontur described its review process as a conveyor — assignments are reviewed by ten developers, each responsible for one task, and the company cut its review time from a month to two weeks. Where hiring happens at scale, assignments do get read.
Take-homes are usually failed not because of the amount of code. Reviewers who publicly describe their criteria name the same mistake: starting to write code before finishing reading the requirements. As one of them puts it, “the requirements explicitly say not to use Linq, and it’s in the code: that’s a hard fail.” The quality criterion everything else comes down to is described at Kontur this way: the best solution is the one the reviewer spends the least time understanding.
Walking through the solution
The market’s answer to generated code is a walkthrough of the solution. hh.ru put this into words back in 2020, long before neural networks: well-designed take-homes always go through a stage where the result is discussed and the solution is reviewed together with the candidate. The walkthrough gives the homework assignment back what it lost: the submitted code can be generated, but explaining it on a call, answering “why this way” and reworking it for a changed requirement cannot.
Nobody has counted how many companies hold such a walkthrough, so it cannot be claimed that it happens “almost always.” What is known is something else: without it, the submitted code says nothing about the candidate.
Internships: where the format lives on
Where an intake draws thousands of applications, manual selection is physically impossible, and the take-home turns into an automated filter. It is the only way to consider such a flow on equal terms for everyone, which is why the format survives in internships.
In selection for Yandex’s backend internship it is the second of four stages: application form, a take-home in Yandex Contest, technical interviews, an interview with the team. Five hours are given for the solution, and the clock starts when the button is pressed. At Ozon, selection for the Route 256 education program is a five-hour contest held on a weekend: each cohort draws three to eight thousand registrations, and results are announced no earlier than ten days after it ends.
Where take-homes are still given
| Who | Is there a take-home | In what form |
|---|---|---|
| Internships, Yandex | Yes, a mandatory stage | A contest: five hours for the solution from the moment of starting, automated checking |
| Route 256, Ozon | Yes, automated selection | A five-hour algorithmic contest on a weekend, 3,000–8,000 registrations per cohort. This is an education program with offers for the best participants, not regular hiring |
| Large-scale hiring of experienced developers: Yandex, T-Bank, Avito | No | Live sections instead of a take-home: language, algorithms, System Design. This was the case before neural networks too |
| Ozon, regular hiring | Possibly | The company notes that for some vacancies — mainly developers and designers — a take-home is sent instead of an interview invitation. Ozon Tech’s role-specific pages describe only live sections |
| Companies that dropped it publicly (SSP SOFT) | No, for middle and senior developers | An hour-long live interview working through a problem. A single statement, not industry practice |
| Small product companies and studios | Often | A homework assignment of several hours; this is also where most of the questionable practices occur |
FAQ
Should I do a take-home for free?
Russian law does not require take-homes to be paid, and most developers are willing to do an unpaid assignment in principle. The question is the scope and what exactly you are being asked to build. A career consultant at hh.ru offers a simple test: “Could the company take your result tomorrow and use it with minimal changes? If so, that’s work, not an assessment.” A two-to-four-hour assignment after the first conversation is normal practice; an assignment that reproduces part of a real product is a reason to discuss payment.
How long should a take-home take?
Both sides converge on two to four hours. In an interviewing.io survey, more than 80% of engineers named four hours as the ceiling, and the most common answer was two. hh.ru frames the benchmark differently but similarly: a take-home should not take longer than one interview, that is, an hour and a half to two hours, and anything over four hours is a reason to talk about payment or a smaller assignment.
Can I use AI to do a take-home?
Explicit bans are rare, and that is exactly why trust in the format collapsed. What matters is not the fact of using it but what comes next: on the follow-up call you will be asked to explain the solution, justify your approach and change the requirements on the fly. A senior frontend developer at Doubletapp states the position things are moving toward: “what matters more to us is that the person did the take-home sensibly and understands what they did, not what tools they used.” The worst outcome is code you can’t explain: it gets you to the very meeting where that comes out.
Do senior developers get take-homes?
In large-scale Russian hiring of experienced developers, almost never: Yandex, T-Bank and Avito build their assessment on live sections. Attributing this to neural networks is wrong — it was the case before them. So far only one small company, SSP SOFT, has publicly dropped take-homes for middle and senior developers citing AI, calling the format “a vestige of the hiring process.” That is a single statement, not industry statistics.
How do I decline a take-home without losing the vacancy?
Decline with a counteroffer. One wording that works: “A take-home that runs over several days doesn’t work for me, but I’m happy to go through any of your problems live — as live coding or a code review, an hour or an hour and a half.” It reads as willingness to take a tougher format, and it offers exactly what take-homes are now being replaced with. Bear in mind that companies rarely offer a choice between a take-home and a live interview: in the interviewing.io survey only 10% of candidates had been given one.
Is it true that only 2 in 10 programmers do take-homes?
No, that is a distorted retelling of a SuperJob study. According to its data, 19% of programmers always complete take-homes, another 54% do so sometimes, and 27% never apply to such vacancies. Almost three quarters agree at least sometimes. The “2 in 10” figure appears if you take only “always” and forget about “sometimes.”
Sources
- SuperJob: how Russians feel about take-home assignments (in Russian) — February–March 2025, 1,400 respondents: among programmers, 19% “always,” 54% “sometimes,” 27% “never” — the worst result of any profession.
- CoderPad, State of Tech Hiring 2026 — February 2026: 12% of recruiters consider the take-home a reliable signal; 52% name solutions generated by AI agents as the main risk.
- interviewing.io: why engineers don’t like take-homes — a survey of around 700 engineers: time thresholds, payment (58% believe take-homes should be paid, 4% had been), only 10% offered a choice of format. The audience is American.
- Anketolog: every second job seeker turns to AI (in Russian) — June 2026, 1,800 people who had looked for work through online platforms: 30% asked a neural network to help with a take-home.
- SKB Kontur: how take-home review works (in Russian) — review as a conveyor: ten developers, code assessment criteria.
- Ozon Tech: how we choose the Route 256 selection tasks (in Russian) — August 2024: cohort size, contest format, timelines.
- Selection for a backend internship at Yandex (in Russian) — July 2024: four stages, five hours for the take-home in Yandex Contest.
- AGIMA: a survey on take-home assignments (in Russian) — October 2023, 214 people in a mixed sample: thresholds in hours. Read with caution — these are judgments of acceptability, not behavior.
The other stages of the funnel are covered in the overview of developer interview stages. What take-homes are being replaced with is covered in the breakdowns of live coding and System Design.