A founder I know sends every engineering candidate a take-home assignment. Four to six hours of work. A real-world problem. The kind of thing that, in theory, reveals exactly how someone thinks and codes.
In practice, here's what happens: roughly half the candidates never submit. The ones who do submit took anywhere from 4 to 14 hours — despite the "4-6 hour" brief. And when the hiring manager asks a follow-up question in the debrief call, half of the finalists can't explain what they wrote.
This is not a niche problem.
The logic made sense when it was invented. Instead of asking candidates to solve LeetCode puzzles under artificial time pressure, give them a real problem and the time to do it properly. No nerves. No tricks. Just actual work.
The problem is the world changed around the format. And the format didn't.
Three things broke it:
Live coding interviews solve the AI-cheating problem but create a new one: they heavily penalise candidates who think out loud differently, have interview anxiety, or don't perform under artificial observation. You end up optimising for "good at being watched coding" — which is a real but relatively minor part of most engineering jobs.
Portfolio reviews are a better signal but only if the candidate has something to show. Junior engineers and career-changers — often the most coachable hires — have thin portfolios. You're also taking their word for what they actually built vs. what the team built.
Generic AI screening tools — the chatbot-sends-you-10-questions variety — solve the scheduling problem but not the signal problem. Static question sets can't probe deeper when an answer doesn't add up. A candidate who says "I led the system design for our recommendation engine" needs to be asked what tradeoffs they made, what broke first, what they'd change. A fixed script never asks that follow-up.
The issue with every alternative is the same: they all measure a proxy. They don't measure whether this person can actually think through a technical problem in dialogue — which is what engineering work actually looks like.
The best technical screens share three properties that neither take-homes nor most alternatives have simultaneously:
This is the gap Ray is built for. Rather than sending a take-home or booking an engineer for a screen, Ray conducts a structured adaptive voice interview — Screening, DSA, or System Design — and produces a debrief report your team can review and act on.
When a candidate says they "optimised the database queries to reduce latency," Ray asks which queries, what the before/after looked like, and what they'd do differently now. That's the conversation where real signal lives — not in the polished code they submitted from their couch at 11pm.
If you're running take-homes right now, the replacement isn't complicated. The sequence that works:
The result is a smaller, higher-quality candidate pool reaching your engineers, with a structured debrief already in hand before the live conversation starts.
Fewer wasted hours. Better signal. And candidates who actually know what's on their resume.
Ray is onboarding early-access teams now. See how adaptive screening works.
Join the Waitlist at Diyam AI →