When a takehome comes back polished, you can't tell what's the candidate and what's Claude. The code stopped being the signal. How they work with AI is the new signal, and Seal is how you see it.
This is what reviewing a takehome looks like today, and what it looks like when you can see the conversation behind the code.
The artifact is polished, and tells you almost nothing. You either distrust it entirely, or burn the first 20 minutes of the onsite re-litigating "did you actually write this?"
Now the takehome is real signal plus interview prep. You know how they think, and exactly which 2-3 moments to dig into live.
Use the assignment you already have. The candidate gets the repo and runs one command: npx @seal-ai/cli init. That's the whole setup.
No lockdown browser. No webcam. No keystroke logging. AI isn't just allowed, it's the point: they work the way they'd work on the job.
The final PR plus the full AI transcript, side by side. See what they asked, what they verified, and where they pushed back.
Monitoring tools assume candidates are cheating. Seal assumes the AI conversation is the work sample, and everyone knows it going in.
Not a score. Not a cheating verdict. The concrete behaviors that separate engineers who drive AI from engineers who paste from it.
"I've spent 16 years interviewing engineers, and takehomes were always my best signal, until LLMs made every submission look the same. But watching how people prompt, verify, and push back on coding agents turned out to be an even better window into how they think than the code ever was. The only thing missing was a way to see it. So I built one."
If you're a founder or engineering leader running takehomes, let's talk for 30 minutes. Worst case, you leave with a better takehome process.
Book a 30-min call