Buried in the coverage of Google's 2026 interview changes is a format almost nobody has trained for. Alongside the AI-assisted round, Google added a code comprehension round: you are handed an existing codebase and asked to read, debug, and optimize real code — not write it from scratch.
Shopify does a version of this. Meta's AI-enabled round leans on it. The pattern is the same everywhere: the blank editor is being replaced by someone else's messy file, and the question shifts from "can you produce this" to "can you understand this."
That is a problem for most candidates, because comprehension is the one interview skill almost no one deliberately practices.
Why companies switched to reading code
The honest reason: it is what the job is. Engineers spend far more time reading code than writing it — joining a team means inheriting a codebase, and AI tools now generate code faster than anyone can review it. The bottleneck moved from production to comprehension.
The classic round tested whether you could generate a correct solution. The comprehension round tests whether you can stand in front of code you did not write, build an accurate mental model of it fast, and find the thing that is wrong. That is the daily job — and it is the skill AI made more valuable, not less.
When an AI can emit 200 lines in four seconds, the scarce skill is the human who can read those 200 lines and say "line 140 has a race condition." The comprehension round is companies hiring directly for that scarce skill.
What the round actually looks like
You get a file, a class, or a small system you have never seen. The interviewer asks you to explain what it does, find a bug, or make it faster. The trap is rushing to a fix before you understand the shape of the thing.
| Weak comprehension | Strong comprehension |
|---|---|
| Starts editing lines immediately | Reads the whole thing first, states what it does |
| Guesses at the bug | Forms a hypothesis, then locates evidence |
| Silent while scrolling | Narrates the mental model out loud |
| Optimizes a random line | Identifies the actual hot path, then optimizes |
| Treats it as a single puzzle | Maps inputs, outputs, and the failure modes |
The interviewer is not watching your typing. They are listening to whether an accurate model of the code is forming in your head — and whether you can say it out loud as it forms.
The skill underneath: reading is an active loop
Beginners read code top to bottom like prose. Strong engineers read it like an investigation — forming questions, checking them against the code, and updating a model. That loop is what the round measures, and it is fully learnable.
Notice that writing code is the last step, not the first. Candidates who invert that order look panicked; candidates who follow it look like senior engineers, because this is literally how senior engineers approach unfamiliar code.
How to prepare
The candidates who struggle with the comprehension round are not worse engineers. They are engineers who only ever practiced the blank-editor version of the interview, while the job — and now the interview — is mostly about understanding code that already exists.
Practice reading. It is the half of the job the old interview never tested, and the half the 2026 interview is built around.
Rubduck lets you practice talking through unfamiliar code with a live AI interviewer that drops you into a problem and probes your mental model out loud — the exact pressure of a comprehension round. You build the habit of narrating your understanding before you reach for the keyboard. Start your free sessions →