X (Twitter) Interview Questions
& Process for Software Engineers

Before anyone at X watches you write code, you have to write about the best code you have already written. The Statement of Exceptional Work goes in alongside your CV: one specific, high-impact thing you personally built, in your own words. It descends from the note Musk sent engineers in November 2022 asking them to summarize their recent commits, and it front-loads the whole process. The first gate is a writing gate. Everything after it belongs to xAI, which runs engineering hiring for the platform now, and the process tends to close inside two to three weeks.

The screen that follows is proctored. Camera and microphone stay on for the full sitting, the clock is what candidates complain about far more than the problems, and some pipelines swap it for a live pairing session on applied object-oriented design: iterators, key-value stores, caches. After the onsite the panel files written feedback and a debrief settles the hire and the level in one conversation. No hiring committee sits above that, though Musk is reported to keep the final sign-off, so an offer is not firm until that lands.

The two things that decide this loop are the ones almost nobody drills: designing a data structure, and keeping it correct once more than one thread touches it. The design round hands you a small stateful component to build from scratch, and your interface is graded as heavily as your implementation. Then it extends: make it thread-safe, then make it survive millions of concurrent queries. That extension is where a candidate with a spotless pattern catalog can still lose the round.

Round counts move with level and team, and reports vary at the edges. Interviewers pick their own questions, so the person you draw shapes the day more than any published rubric would suggest. Reports also disagree about whether the behavioral portion is a round of its own or a conversation folded into the day. Either way there is no values list to rehearse against, and one honest, specific answer to why you want to be here does more work than a stack of STAR templates.

CodingCore

Practical, medium-centered data-structure and algorithm problems, graded on clean, working code over trick-spotting

Data-structure designDesign

Build a small stateful component from scratch, with a clean API, correct edge cases, and attention to thread safety

System designScale

Design a large-scale service end to end, reasoning about access patterns, fan-out, partitioning, and failure

BehavioralCulture

Ownership, collaboration, and genuine interest in the product, anchored in how the team builds and operates

The coding rounds are medium-dominant with a hard tail, and they are rarely lost on difficulty. The score comes from whether the thing you wrote would hold up in a running service: an interface you can defend, edge cases you found before the interviewer pointed at them, and shared state that does not corrupt when two threads reach it at once. Simple and correct beats clever and broken, and most people who lose here lose by reaching for the clever version.

Around those two sit the ordinary families: arrays, strings, and hashing; trees and graphs; heaps, top-K, and intervals; and dynamic programming. DP is there for breadth rather than specialization, so expect to see it once and do not build a month of prep around it. Candidates who do well state their complexity without being asked and walk the edge cases before anyone goes looking for them.

The home timeline is the canonical system design question in the industry, and this is the company that produced it. That does not mean you are guaranteed to get it. Fan-out on write is the answer everyone arrives with, and it breaks the moment an account with ten million followers posts. The real system is a hybrid: timelines precomputed for ordinary accounts, the large ones merged in at read time. Knowing roughly where that threshold sits and being able to argue for it counts for more than naming another queue. The company open-sources the recommendation code behind that timeline as the-algorithm on GitHub.

The other worked designs sit on the same platform: trending topics as top-K over a firehose, a rate limiter, notifications and fan-out, and real-time search across billions of posts. The rate limiter earns extra attention, because it turns up in both eras of this company and because the public API here has long rate-limited with GCRA, a leaky-bucket variant that tracks one timestamp per key rather than a counter.

The strong answers start at the access patterns and the consistency requirements, then get concrete about fan-out, partitioning, and what happens when a shard dies. A tour of technologies with no trade-off attached to any of them reads as someone who has read about distributed systems rather than run one.

An X offer no longer involves public stock. Twitter was taken private in the 2022 buyout, its TWTR ticker was delisted, and the company folded into xAI in a 2025 all-stock deal, so the equity in an offer today is private xAI stock rather than anything you can price on a public market. That retires everything ever written about the old public-Twitter RSUs and replaces it with the usual private-company questions: what a share is worth, and when it turns into money.

Historical Twitter-era average total comp by level, per Levels.fyi (US). Read this as a Twitter salary reference, not a current X offer: most reports predate the 2022 take-private, Levels.fyi still lists the delisted TWTR ticker and public RSUs, and X is now private under xAI. The aggregate median was about $223K; Senior Staff and above sit higher on very thin samples.

Reports from the Twitter era describe no annual bonus and no structured refresher program, and a private company under new ownership may run things differently, so ask rather than assume. If the initial grant and the sign-on carry the whole offer, that is exactly where a counter belongs.

The lever that works is a written offer from one of the labs bidding for the same small pool of engineers, and that competition is expensive enough to move a number here. Get the grant in specifics before you sign: the instrument, the share count, the valuation the price is struck against, the vesting shape, and any liquidity constraint on your shares, because a private grant is only worth what a future tender or event lets you sell it for.