You should be nervous about the questions, not the software.

Know when and what

The invitation says the round, roughly what you will be asked to do, and the time where you are — not UTC, and not the company's zone — with a calendar file attached.

A reminder that helps

The day before and an hour before, with the joining link in it — and, if it is your first one, the download and what to expect.

An answer, either way

When a decision is made you hear the same day, whichever way it went. Nobody is left to find out by silence.

Write and run real code

A real repository and a real shell. Any language, any toolchain — it is your machine, so whatever works there works here.

Show the thing running

Start your app and the interviewer can open it, without a deployment or a tunnel you had to set up.

Use your own agent, if the round allows it

Sign in to an agent already on your machine and work the way you actually work. The interviewer sees what you asked it and what it did — which is the point, not a gotcha. Some companies turn it off for a round; you will know which before you start.

It never reads your screen, your keystrokes, your files or what is in any other application. It cannot stop anything from running. Declining is recorded as declining — not as something worse — and what it does find reaches the interviewer as context rather than a score. You are told all of this in the room, before anything runs.

Start a mock session and invite a friend — another candidate, or somebody willing to play interviewer. They join with a meeting ID and a passcode, in the same room, with the same repository, terminal and agent as the real thing. No company involved and nothing recorded against you.