Practise Meta’s AI-enabled coding interview in a real codebase.

The round runs in CoderPad. There is a file tree, an editor and a chat panel where you can pick between several AI models. The assistant can explain code and suggest changes. You stay responsible for what goes into the files and for showing that it works.

Instead of one function to write, you get a small project: several files, classes that already exist and tests. Candidates describe a single extended problem with checkpoints, where each stage builds on the last.

1. 01

   Fix what is broken

   The session usually opens with a bug in code you did not write: a wrong condition, an off-by-one, a bad type conversion. You have to find the entry point and follow the data before you can change anything.

2. 02

   Build the core

   Next comes a real feature on top of the existing classes, often a hundred lines or more. The AI can draft pieces. You decide the design and check its output against the tests.

3. 03

   Make it hold up

   The last stage raises the bar: larger inputs, a slow path that has to go, or an edge case the first version missed.

• How fast you get oriented in code you have never seen
• Whether you verify: running tests, reading what the assistant wrote and catching it when it is wrong
• Changes that fit the existing design instead of rewriting it
• Clear communication about what you are doing and why

Using the AI is allowed and expected. Leaning on it without checking is what costs candidates the round.

Algorithm drills train you to write a known pattern into an empty file. The AI-enabled round starts from the other end: a codebase that already exists, conventions you have to follow and a bug you must find before you can build anything. The first ten minutes, spent reading, shape everything after them.

StackYard recreates the conditions of this round on real code. Each task is a ticket in a repository you have not seen, opened in the browser with everything already set up.

• Real tickets from real open source repositories and realistic services, in Python and TypeScript
• A browser IDE with a terminal and the project’s own test suite, nothing to install
• Pair, an AI assistant beside the editor that reads the code you are working on
• Example cases to run while you work, and hidden checks when you submit, so a pass means the fix holds
• A time budget on every task, from 10 to 45 minutes

A good session for this round: pick a hard multi-module service, spend ten minutes mapping it before you write code, ask Pair to explain anything unclear, then fix, verify and say out loud why your change is right.

1. 01

   Week one: read before you write

   Two or three easy tasks a day. Before editing, write down where you think the bug is and why, then check yourself against the tests.

2. 02

   Week two: work with the assistant

   Move to medium tasks. Ask Pair to explain a module, then confirm what it said in the code. Practise spotting and rejecting a wrong suggestion.

3. 03

   Week three: full sessions

   Hard, multi-module tasks inside their 45-minute budget, talking through your reasoning as if an interviewer were listening.