A growing number of companies hand you the tools and watch how you use them. Same 45 minutes, completely different scoring, and a very obvious gap between people who direct a model and people who follow one.
Written after two AI-permitted rounds · Updated September 2026 · 8 min read
AI tools: allowed in this round
45:00
!
The short version: when AI is allowed, the code stops being the signal. What gets scored is whether you had a plan before you prompted, whether you reviewed what came back, and whether you can defend the result under questioning. Prompting fast and pasting fast is the fastest way to fail an open-tools round.
This is about rounds where AI is explicitly permitted. When it isn't, using it is cheating, whatever a tool's marketing says.
The round where my instincts were wrong
I'd trained myself to hide the tools. Then they handed me one.
Every piece of advice I'd absorbed said: close everything, prove it's all you. Then an interviewer said, "use whatever you'd use on a normal workday, just talk me through it." I froze for a second, which is a funny way to fail a round designed to be easier.
What I learned over the next forty minutes is that open-tools rounds are not a relaxation of the standard. They move the standard somewhere harder to fake.
When the code is free, the only thing left to evaluate is your judgement about it.
What actually gets scored
Four things, none of them typing.
A plan before the prompt
"Here's my approach; I'll use the tool for the boilerplate." Said out loud, before anything is generated.
Review of what came back
Reading the output critically, out loud, including what you'd change.
Knowing when not to use it
Writing the core logic yourself is a signal, not a weakness.
Defending the result
Every line is yours the moment you submit it.
Two candidates, same tools
The difference shows up in one question.
How to practise for open-tools rounds
Narrating is the skill.
Practising this alone is awkward, because the natural way to use a tool at home is silently. So I made the narration the point. In my nightly sessions with Phantom Code AI I'd state my plan out loud first, then capture the problem and compare its approach and complexity against mine, saying what I'd keep and what I'd change. It covers coding, system design, SQL and behavioral rounds, so the same habit carried across the whole loop, including the closed-tools rounds where I practised without it entirely.
What it can't do is play the interviewer who interrupts you at minute twelve to ask why. A friend did that once a week, and those sessions were where the narration actually became natural.
The closed-tools rounds still exist
And the rules there are absolute.
Most interviews are still no-tools, and in those, hidden assistance is cheating regardless of how a product markets itself. Detection keeps improving on assessment platforms, and the follow-up question exposes borrowed answers faster than any software does. If the round is closed, close everything.
Ask which kind of round it is. Ten seconds of clarification beats forty-five minutes of guessing.
Common questions
Short answers.
Do companies allow AI now?
What's scored when it's allowed?
Should I use it if I'm faster alone?
What if they don't say?
The tools stopped being the secret
For two years the anxious question was whether you could get away with using AI. The more interesting question now is whether you can be trusted with it, and that one you answer out loud.
Next practice session, narrate the whole thing as if someone is listening. It's harder than it sounds and it's the entire skill.
— Written after being told to use whatever I'd normally use
Have you had an AI-permitted round yet?
I want to know how common this is getting.
Some third-party reviews of these tools are published by competing products, so verify prices on official sites before buying.
Top comments (0)