Courses › AI-Assisted Coding › The New Round

Introduction

What AI-assisted coding interviews look like and what they actually evaluate.

A growing number of companies now run coding interviews where candidates may — or are expected to — use an AI coding assistant. These rounds reflect how engineers actually work today. But they don't make the interview easier: they change what's being measured.

What these rounds look like

Formats vary, but common ones include:

  • Build a feature in an existing codebase — add an endpoint, a CLI command or a module to a small repo with tests, in 45–90 minutes.
  • Build something from a spec — a small app or tool from a one-page spec, often with deliberate ambiguities.
  • Debug — fix several planted bugs in an unfamiliar codebase and explain each root cause.
  • Multi-level problems — like classic practical rounds (an in-memory database built in stages), but with AI help allowed.
  • Data and integration tasks — process a messy file, analyze logs, or integrate a third-party API from its docs.

You'll usually share your screen and narrate. Some companies provide a specific tool; others let you use your own setup.

What's being evaluated

When an assistant can write code quickly, typing speed and syntax recall stop mattering. What remains is judgment:

[!WARNING] The most common failure Pasting the whole problem into the assistant, accepting a large blob of code, and then spending the rest of the interview debugging something you don't understand. Interviewers read this as an inability to engineer without the tool.

What doesn't change

You still need strong fundamentals. To evaluate the assistant's output, you have to know what correct, efficient, idiomatic code looks like — and spot when it isn't. Candidates with weak fundamentals tend to do *worse* in AI-assisted rounds, because they can't tell good suggestions from bad ones.

What's in this track

The core skills

Round types

Tools & pitfalls

Then practice with the AI-assisted questions in the question library.