For a decade the take-home assignment was the default way to screen engineers: here is a problem, spend a weekend on it, send it back. It was always a compromise, and two things have now broken it badly enough that the best teams are moving on. One is that it never measured what it claimed to. The other is that AI quietly removed whatever signal was left.
It selected for free time as much as skill
An unpaid eight-hour assignment does not test ability evenly. It tests who has eight unclaimed hours. A senior engineer with a current job, a family, and other interviews in flight is at a structural disadvantage against someone with an empty calendar, regardless of who is better at the actual work. Teams that lean on take-homes systematically filter toward the candidates with the least going on, which is rarely the same group as the strongest engineers. It also quietly pushes away senior people who simply decline to do unpaid work, and those are often the ones you most wanted.
AI made the take-home impossible to grade
Even setting fairness aside, the take-home no longer tells you what you need to know. A current model will produce a clean, well-structured, tested solution to almost any standard take-home prompt in minutes. You can no longer tell from the artifact whether the candidate reasoned through the problem or pasted it. Grading the output now mostly measures how well someone edited an AI’s answer, which is a real skill but not the one most rubrics were written for. The take-home has become a test you cannot score honestly.
What serious teams do instead
The better approaches all share one trait: they watch the engineer think, in real conditions, on work that resembles the actual job. Some teams run a short paid trial, a few days on a real, scoped piece of work, paid at a fair rate. Some do a structured pairing session on a real bug or feature in a sandbox of their own codebase, where the point is the conversation and the decisions, not a finished product. Some ask the candidate to walk through a pull request they already wrote and defend the choices. Each of these is harder to fake with a model, fairer to busy seniors, and far more predictive than a polished weekend submission.
Why this matters more for remote hiring
If you are hiring remote engineers, especially across borders, the evaluation problem is sharper because you have fewer informal signals to fall back on. That is exactly why the live, conversational, work-like formats win: they show you how the person reasons, communicates, and handles being wrong, which is most of what makes a remote senior engineer effective. A take-home shows you none of that. It shows you a file.
We pre-vet every engineer with a live technical conversation and a real code review before we ever introduce them, so you are not starting your evaluation from zero. See how the vetting works.