Figma Interview Questions
& Process for Software Engineers

Silence costs you here even when you land on the right answer. Figma's interviewers pair with you rather than watch you: they will argue about a variable name or drop an edge case in halfway through, and how you absorb that goes into the write-up next to whether the code ran. Someone who narrates a slower path and lets the interviewer into it usually scores above someone who reaches the optimal answer without saying much. The loop around all of that is ordinary enough: a recruiter screen, a technical phone screen with a working coding problem, then a virtual onsite that mixes coding, system design, and behavioral rounds.

After the onsite the panel files written feedback, and one debrief settles the hire and the level together before anything moves to the offer, which is why a yes you are happy with and a level you are not can arrive in the same call. The hiring manager holds final authority when the room does not converge. That makes the whole thing feel less like clearing a company-wide bar and more like being hired by one particular team.

The onsite is one block that runs coding, system design shaped by Figma's own product, and a behavioral hour about how you work and how much you care about what you ship. The split moves with level, team, and track: the frontend and rendering loops trade part of the distributed-systems conversation for the browser paint path, and they go deeper than most companies' frontend rounds do.

The behavioral hour usually turns into a walk through a project you owned, and the questions keep landing on why rather than what. Answers that carry numbers hold up: the latency before and after, the share of a machine you handed back. Generic ownership language gets pushed on until it either produces a specific or falls over.

CodingAlgorithms

Medium-centered algorithmic problems where clean, working code and clear communication matter as much as the optimal answer, spanning arrays and hashing, trees and graphs, dynamic programming, and intervals

System designProduct

Collaboration-shaped design questions drawn from what Figma actually builds: real-time multiplayer editing, live presence and cursors, the canvas rendering path, file storage and version history, and comments

Behavioral & craftValues

A conversation about how you work, how you handle ambiguity, and how much you care about the details of what you ship, grounded in genuine interest in Figma and its products

The algorithms themselves are ordinary and sit around medium: arrays, strings and hashing, hash-map and data-structure design, trees and graphs, dynamic programming, and heaps and intervals. They are just rarely posed as themselves. Implement undo and redo over a document model, merge the styled ranges inside a block of text, walk a layer tree. The algorithm underneath is one you have seen before, and the interviewer is watching how you model the thing sitting on top of it.

The code runs in this round, so the tests pass or they do not, and passing them is where the evaluation starts rather than where it ends. State the invariant before anyone asks for it, which on an undo problem means saying out loud that a fresh edit clears the redo stack, and give the complexity without being prompted. A readable solution you can defend under a follow-up beats a clever one you cannot explain, because the follow-up is coming.

The anchor problem is real-time multiplayer editing, and the prep answer everyone brings to it is CRDTs and operational transforms. Figma looked at both and shipped neither, and it published the reasoning: operational transforms carried more complexity than the problem needed, and a true CRDT carries coordination overhead built for peer-to-peer systems that a central server makes unnecessary. What runs instead is one authoritative server process per document holding the latest value any client has sent for a given property, with the last write winning at that property boundary. Reciting the textbook comparison here puts you on the wrong side of a decision the company already made in public.

From there the round branches. Live presence and cursors looks like the same problem and is not, because a cursor position is disposable and an edit is not, and the answer that scores treats them as two channels with two different guarantees. The canvas rendering path comes with a sixteen-millisecond frame budget and an engine written in C++ and compiled to WebAssembly. File storage and version history becomes a question about snapshots against operation logs, and comments with their notifications sit outside document sync altogether, in relational storage, on a consistency story of their own.

Graders are after the reasoning more than a particular diagram. Naming the consistency model you picked, saying what breaks when a partition heals or when two hundred cursors land on one file, and tying it back to what a real user feels will carry a round further than an architecture that is technically fine and drawn in silence.

An offer is base salary, equity, and sometimes a sign-on, and since the 2025 listing the equity is ordinary public-company RSUs on a four-year schedule, a quarter at the one-year cliff and monthly after that. The grant is written as a dollar figure and converted into shares at the price on the day it is cut, so the number in the letter and the number in your account start drifting apart the moment you sign.

That is a good reason to give base more attention than it usually gets. Equity holds the real upside and all of the volatility at the same time, and a stock that has been above $140 and below $20 since it listed makes that component a moving target rather than a promise. Treating the whole package as one number is the only way to compare it against anything else on your desk.

Average total compensation by level, per Levels.fyi (US, as of Aug 2026); self-reported and rounded. Figma labels these levels by number only; the Entry, Senior, and Staff mappings are the market-standard reading by years of experience. Since the 2025 listing the equity is public RSUs shown at an annualized value. The overall median is about $430K, and L5 reports near $733K on a thin sample.

The conversation belongs after the decision comes back positive and while the offer is still being written. A competing offer moves more than anything else you can put on the table. Without one, the lever is the level: hire and level were settled in the same debrief, so a clear case for your scope is a case about the whole band. Asking where the top of that band sits costs nothing and gets skipped constantly.