Meta's AI-enabled coding round
A round with an AI assistant in the editor is a different test from an algorithm puzzle. The practice version here gives you a small codebase and scores whether you understood it, checked what the assistant wrote and can explain the result.
What is known about the real round
One coding round in the onsite is reported to be a 60-minute AI-enabled round with an assistant built into the editor.
Candidates and prep sites report that Meta began piloting this round in October 2025 and that it replaces one of the two coding rounds. The assistant is the one inside the interview environment; your own tools are not used. Reports disagree on which models are offered. We could not find a public Meta document describing the round, so confirm with your recruiter.
Formats change between hiring cycles. Ask your recruiter what your round will be, and see the policy tracker for other companies.
How the practice round runs
Placed runs its own version of this format with a voice interviewer. The codebase and the tasks are Placed's, not Meta's.
- Step 1Find the reported bug
You get a small multi-file project and a bug report. In the Placed version the assistant is closed for this phase, so you read the code and the failing behaviour yourself and say what you think is wrong before you change anything.
- Step 2Implement the feature
The assistant opens. You add a feature to the code you have just read. What you ask the assistant for, and what you do with the answer, is on the record.
- Step 3Make it scale
The input grows and the first version is too slow. You say where the time goes, pick a fix, and show it still returns the right answer.
What gets scored
The report scores 4 things from 0 to 10: problem solving, code quality, verification, communication. On Pro, each score comes with the moment in your session it is based on. How grading works.
Where people lose the round
- Pasting the whole task into the assistant and accepting what comes back.
Ask for one piece at a time, and say what you expect before you read the answer. Then compare.
- Not running anything until the end.
Run the tests after each change. An interviewer watching you verify is watching the thing the round exists to measure.
- Going quiet while reading code.
Narrate what each file is for as you open it. Silence reads as being lost, even when you are not.
- Being unable to explain a line the assistant wrote.
If you cannot say why a line is there, rewrite it or delete it. You own everything you submit.
Three sessions before the real one
- Sit one three-phase round cold and read the report. Look at the Verification score first.
- Do a debug-only round with no assistant, so reading unfamiliar code stops being the slow part.
- Sit the three-phase round again a few days later and compare the two reports criterion by criterion.
Placed is not affiliated with Meta. Company names are used to describe publicly reported interview formats.