Interviews · AI

Google's code comprehension round: what it is and how to prepare

21 Nov 2026 · By Akshay, software engineer, India

Google is piloting an interview round where you work inside an existing codebase, with AI allowed. What's known so far, what it tests, and a practical way to prepare for code-reading rounds.

For years, the coding round meant one thing: solve an algorithm problem on a blank screen. That’s starting to change. Google has confirmed it is piloting a code comprehension round, and several other companies now run rounds where you work with code someone else wrote, sometimes with an AI assistant open next to you.

What we know so far

  • The format. Instead of writing a solution from scratch, you work inside an existing codebase: find bugs, improve performance, and refine code that’s already there.
  • AI is allowed in the pilot. A Google spokesperson confirmed to Business Insider that candidates in the pilot use Gemini as their assistant. Interviewers look at how well you use it: how you prompt, how you check what it gives you, and how you debug when it’s wrong.
  • Who it’s for. The pilot is for junior and mid-level software engineering roles on some US teams, from the second half of 2026, with plans to expand if it goes well.
  • Algorithms aren’t gone. Reports say the data structures and algorithms rounds stay, without AI.

Google isn’t alone. Meta reportedly added an AI-enabled coding round in October 2025, and some companies let candidates use AI assistants in live coding. The details differ by company and change often, so treat any single report as a snapshot.

Why this is happening

When AI can write a first draft of almost any function, the scarce skill is reading: understanding what code does, spotting where it breaks, and deciding whether a change is safe to merge. That’s what these rounds test.

What they actually test

  1. Reading code you didn’t write. Following the flow through several functions and files without running it.
  2. Predicting behaviour. What prints, what’s returned, what happens on an empty list or a slow network.
  3. Finding the bug. Not just “it fails”, but why and where.
  4. Improving it. Making it faster or clearer without breaking anything else.
  5. Working with AI. Asking precise questions, checking the answer instead of trusting it, and noticing when it’s confidently wrong.
  6. Thinking aloud. Explaining what you’re checking and why, as you go.

How to prepare

  • Read before you write, every day. Ten minutes of “what does this print?” builds the habit faster than another hour of new problems. Start with our Code Reading drills.
  • Trace by hand. For async code especially, write down the call stack and the queues step by step. The event loop lesson shows how.
  • Read real code. Pick a small open-source project, read one file a day, and sum up what it does in two lines.
  • Review, don’t just write. For each change in your own projects, write the review comment you’d want to get: what could break, what’s missing a test.
  • Practise with AI, then check it. Ask an AI assistant to explain a function, then verify each claim by running the code. Note where it was wrong.
  • Keep the fundamentals sharp. Complexity, data structures and the language’s rules (scope, closures, this, promises) are what let you read fast. Our interview questions cover the ones that come up most.

The short version

The round is new, but the skill isn’t: careful reading, honest checking, and clear explanation. Start practising it now, and the format won’t surprise you.

Sources: Business Insider’s report, via Yahoo Finance · Entrepreneur · A tracker of companies that allow AI in interviews

Practice, don't just read

Free lessons with videos, code-reading drills and interview questions.