You'll talk through your experience with an AI screener, then build something real with an assistant beside you. This page covers what happens, what's recorded, and your rights, so nothing on the day is a surprise.
Type or speak your answers. Switch whenever you like.
Always
A person makes the decision.
Your path
Three steps, and you're told about each before it starts.
1
Apply
Send your details and a resume. An AI reads it against the role and a recruiter decides what happens next. You get an email receipt.
2
The AI screen
A conversation about your experience that follows the same plan for every candidate. Type or speak, with a whiteboard or code when a question calls for it.
3
The coding round
A real project in a real editor with an AI assistant. The company decides how much the assistant can do, and the editor tells you.
1Applying
Your resume is read for what it says.
Before you send anything, the form tells you an AI will read your resume and a person will decide. The notice you saw is saved with your application.
Every judgement quotes your resume. If the AI can't point to a line you wrote, the judgement is thrown out.
Nobody is turned down by software. A recruiter moves you on or declines you. A decline needs their written reason.
You hear back by email, whatever the outcome. The recruiter's internal reason is never in it.
2The AI screen
A conversation, at a time that suits you.
You'll see a consent notice, then a practice question that isn't recorded, before the first real one. The screener asks about your work and follows up when an answer is vague. Some questions open a whiteboard or a small code workspace. Try the toggles below.
Section 3 of 4 · System designplan v3 · approved
ScreenerLet's design the notification fan-out you described. Sketch the pieces on the board as you go.
●Listening…
saved with each answer
Speak or type
Where the company allows voice, you can talk instead of typing and switch back at any point. You're told whether audio is kept before you consent and again before the microphone opens.
Stop whenever you like
You can end the screen at any point. It's recorded as you having stopped, not as a failure.
Your tests count
In a code question you write your own tests. How you chose to check your work is part of the answer.
3The coding round
An editor you already know how to use.
Monaco, the editor inside VS Code, with a file tree, terminal, output panel and live preview. The project comes set up — React, Next.js, Expo or a Node service — with tests already in place.
Inline completions you accept with Tab or ignore.
Chat that sees the files you tag with @, and a model switcher.
Agent mode that waits for you to allow or cancel each edit.
Plan, Build and Review — use them however you like; none is required.
typed · 9 linesaccepted from AI · 4 linesedited inside an AI region · 1 line
The clock
The round is timed, and you can always see the timer.
The company sets the length, usually 30 to 90 minutes. The clock starts when you consent, not when the invite was sent, so reading this page first costs you nothing.
You can finish early
Submit and finish hands over your workspace exactly as you left it. You're asked to confirm, because it can't be undone.
When time runs out
The session ends and your work is saved as it stood. Running out of time isn't a failure; it's recorded as what happened.
Your interviewer may ask you to finish
If the round has gone as far as it needs to, they may ask you to submit early. Nobody can end it from outside your editor.
Nothing after the end
Once the round ends the workspace is frozen. No later edit can reach the report.
What the AI can do
It depends on the interview. The editor always tells you.
The company picks one of two modes when it sets up the round, and a badge in the editor shows which one you're in. It stays the same for the whole round: what the assistant can do when you start is what it can do when you finish.
unguarded · full assistant
✓
ChatSyntax, concepts, finding your way round the codebase
available
✓
Full solutions in chatGuarded: refused, redirected to a conceptual hint
available
✓
Inline completionsAccept or reject — each one logged
available
✓
Agent modeReads, edits and runs tests behind Allow / Cancel
available
✓
Model switchThe configured models, plus Auto
available
What is recorded
Your work, in full. Not you.
Recorded
Your resume, and what was read from it.
The screen's transcript, with any drawing or code you made. Audio only if the company keeps it, and you're told first.
In the coding round: your code as it changes, every prompt and response, and completions you accepted or dismissed.
Runs, test results and terminal commands.
Focus changes and pastes from outside the editor, shown to a person as context and never auto-flagged. Some companies switch these off, and your editor badge says so.
Never recorded or analysed
Your face, expressions or emotion. There is no camera-based proctoring.
Your screen outside the editor tab.
Anything after the session ends.
Your personal details in any URL. Invite links are signed tokens.
Anything about you beyond your work. The assessment is told to judge only what you said and built.
What a strong session looks like
Work the way you would with a fast, fallible colleague.
These are the habits the assessment rewards. None of them are tricks.
◎
Read before you prompt
Look at the tests and the code first. A minute of thinking shows up as time to first prompt.
@
Give it context
Tag the files. Say what you expect and what the error is. Specific prompts beat short ones.
✓
Check what you accept
Read AI output before moving on. Big blocks accepted and never tested are flagged for a person to look at.
▶
Run the tests
Running tests after accepting a change is verification — the clearest signal there is.
↺
Fix, don't re-roll
When the AI is wrong, correcting it yourself or with a sharper prompt counts as recovery.
“
Say what you're doing
Communication is scored. Explain your trade-offs in the screen and in the round.
Your rights
A person decides. You can see why.
A human makes the decision
No score, signal or model can reject you. Every outcome is recorded against a named person.
Consent before anything starts
You see the notice and agree to it before a screen or a round begins. Decline, and nothing is recorded.
The link is yours
Your invite works only for your email address. If it's forwarded, consent is refused, so nothing about your interview can be filed under someone else.
Your report, on request
Ask the company for your report, including the scores and the written reasoning behind them.
Deletion, on request
Ask for your data to be deleted and an admin can erase your profile, application, every round and every recording. The log keeps only that a deletion happened, and when.
A limit on how long it's kept
Companies set a retention period, after which recordings and declined applications are erased automatically.
Questions
Before the day.
Am I talking to a person in the screen?
No, and it says so up front. The screener is an AI that follows a plan the company approved. A person reads the transcript and decides what happens next.
Can I say something to the AI to get a better score?
No. Your answers are treated as answers, not instructions, and the plan and scoring live on the server where nothing you type can reach them. Every score has to cite something you actually said.
Will using the AI a lot count against me?
Not on its own. The share of AI-written code is one signal among several, read next to how you checked and corrected it. Accepting a large block and shipping it untested is what gets flagged.
What if I don't use the AI at all?
That's allowed. But in unguarded rounds the company usually wants to see how you work with it, so ignoring it may leave AI fluency unassessed.
Can I switch tabs to read documentation?
Yes. Focus changes are shown to the interviewer as context, the way they'd notice you reading docs on site. Nothing is automatically held against you.
The link says my email is wrong.
The invite only works for the address it was sent to. Use that address, or ask the recruiter to send the invite again to one you can use.
What do I need?
A current Chrome, Edge, Firefox or Safari on a laptop, and a microphone if you'd like to speak in the screen. Nothing to install: the project and its tests run in a sandbox we host.
