Courses › AI-Assisted Coding › The New Round

How to Prepare for AI Rounds

What to practice, a two-week plan, realistic mock setups and reviewing your recordings.

Using an AI assistant every day at work doesn't automatically prepare you for an AI-assisted interview. In an interview, you're timed, you're watched, the codebase is new to you, and you have to explain every decision while you make it. Those are skills you can practice. This page gives you a plan.

What to practice

The last row matters: most loops still include AI-free rounds, and you need strong fundamentals to judge the assistant's output. Keep your coding practice going alongside this plan.

A two-week plan

If you have less time, do days 3–4, one debugging session and two full mocks.

Setting up a realistic mock

• Match the tools. If the company provides an environment (often a browser IDE with a chat assistant), practice with a chat-style assistant rather than your usual agent with full repository access. If you bring your own tools, practice in exactly the setup you'll use. See Tools & Context.
• Use code you don't know. Practicing in your own project hides the hardest part: orienting.
• Set a strict timer: 45, 60 or 90 minutes, matching the real round.
• Narrate out loud, and record the screen and audio. It feels awkward, but the recording is the most useful thing you'll get.
• Get an interviewer if you can. A friend can play the interviewer: answering clarifying questions, adding a requirement halfway through, and asking "why did you accept that?"

Good sources of practice tasks:

• Issues labeled "good first issue" in small, well-tested open-source projects
• Classic practical-round prompts: an in-memory database, a bank, a file store (see Practical Multi-Part Rounds)
• The AI-assisted questions in the question library

Reviewing your recording

Watch it at double speed and score yourself honestly:

Pick one or two warning signs and focus on them in the next mock. Fixing one habit at a time works better than trying to fix everything.

[!TIP] Build a prompt library While practicing, save prompts that worked well: "summarize this repository's architecture and where X is handled", "write tests from this spec, not from the implementation", "change only this function, and keep its behavior identical". In the interview you'll write them faster and get useful output on the first try.

The day before

• Confirm the rules for each round with your recruiter (see Where AI Is Allowed).
• If you're using your own setup, check that the assistant is signed in, the model you want is available, and screen sharing works.
• Re-read Common Mistakes. Most failures come from the same ten habits.
• Do one short warm-up task, not a full mock, so you go in confident rather than tired.

[!TIP] Related Introduction · A Winning Workflow · Reviewing AI Code