xAI interview, now SpaceXAI, and the statement before it.

Published accounts of xAI's process disagree more than most, loops vary by team, and candidates say the process changes often enough that someone else's experience may not describe yours. The stages below are the ones more than one account reports, and they are Grok's; Cursor roles run the separate loop described under the tracks. Timing is the least predictable part: replies range from hours to weeks, and one detailed account runs from an application in late June to an offer letter in late August, with the first interview ten days in and the second six weeks after that. Candidates also report interviewers running late across multiple rounds, and roles being withdrawn part-way through when hiring plans change.

Grok Recruiter is the recorded step in xAI's AI tutor and multilingual loops: no interviewer, about five questions answered one at a time to camera, your audio transcribed and read afterwards. Candidates who sat it in 2026 report the shape below rather than the exact wording, which changes by role and language. It is not part of the engineering loop; core engineering candidates go from the recruiter screen to a hiring manager.

• About five questions, one at a time, recorded: Each question appears, you answer to camera, and you move on; there is no follow-up and no chance to ask what was meant. Answer in full sentences with the point first, then one example, and stop. A minute to ninety seconds each; silence and restarts are transcribed too.
• For multilingual roles, a split between languages: Reported as three questions in English and two in the other language. Do not switch mid-answer: the transcription is graded per language, and a sentence that drifts back to English in the other-language answer reads as a gap.
• Translate a supplied passage: A short text at roughly B2 level, rendered into the other language. Accuracy over flourish: keep the sentence structure where the target language allows it, and do not paraphrase away a clause you are unsure of.
• Explain a grammar point simply: The tutor test proper: pick a rule a learner gets wrong, state it in one sentence, give one correct and one incorrect example, and say why. A teacher's answer, not a linguist's.
• It transcribes you, so speak for the transcript: Slower than feels natural, every word finished, no filler. Candidates who sat it say a clean slow answer scores better than a fast one that has to be untangled, because what is graded is the text of what you said.

If your posting is engineering, data centre or operations, none of this applies to you: those loops go straight from the recruiter screen to a person. The tutor loop is lighter than those, and candidates who have sat both say so.

xAI puts most of its judgment into technical rounds and one piece of writing, so what it rewards is easiest to read from what it chose to ask for first.

• Provable personal ownership. The statement asks what you did rather than what your team shipped, and the deep dive exists to check the answer
• Difficulty over breadth. The prompt asks for your most complex work, so one hard project beats a list of competent ones
• Evidence. Shipped systems, benchmarks, users, cost or reliability numbers carry the statement where adjectives do not
• Operating rather than reciting. Candidates in the technical rounds describe scenarios rather than definitions, and the grading is on the order you would do things in and why
• Adaptability. The process itself changes between teams and months, and candidates who arrived with a memorised script report it fitting badly

The behavioural ground at xAI is narrower than most, and almost all of it traces back to the statement you submitted. Prepare to defend that project rather than to assemble a broad story bank. The scenario questions below are the shape reported for operations and data centre loops.

Levels.fyi for software engineers at xAI in the United States, read on 9 September 2026, where the Level 4 median is $640K made up of roughly $240K base and $400K a year in stock. Levels.fyi quotes $660K as the median in one place and $640K in another, and the highest reported package is $1.2M. The sample is small and concentrated at one level, so treat these as an order of magnitude rather than a band structure. Stock is the larger half of these numbers. The figures were reported before SpaceX listed on the Nasdaq in June 2026, so the equity behind newer offers is in a public company and priced daily, while what is vested and when still depends on the grant. Tutor, operations and data centre roles are paid on separate scales and are not covered by this figure.

1. Write the statement first and give it real time. It is read by engineers before any call, it decides whether there is a call, and it becomes the spine of the deep dive. One project, the hardest constraint in it, the decisions that were yours, and numbers.
2. Cut team credit out of the statement. The reported failure is a strong project described in the first person plural, which leaves a reviewer unable to tell what you did.
3. Do not overclaim your level. Reviewers compare the seniority you imply against the detail you give, and the mismatch is more damaging than a smaller claim would have been.
4. Prepare to defend every line of it out loud. Rehearse the deep dive as an interview rather than rereading the document, because the questions come at the decisions rather than the description.
5. For operations and data centre roles, rehearse incidents rather than topics. Take a network failure, a dead disk, a GPU node that will not come up, and talk through what you check first, second and third, and what you would do if the first answer were wrong.
6. For the recorded tutor step, speak slowly and cleanly. It transcribes what you say, so a slower answer with no mistakes reads better than a fast one that has to be untangled.
7. Expect the process to move oddly, and keep other applications alive. Late interviewers, gaps of weeks between rounds, and roles withdrawn mid-process are all reported often enough to plan around. A delay is not a decision, and a rejection will probably arrive without a reason, because recruiters cite company policy for not giving one.
8. For a Cursor role, use the product for real before you apply, and prepare for the work trial as the round that decides: up to two days in the real codebase with the editor allowed, then a review where every line you kept has to be yours to explain.
9. Read recent candidate threads before you commit weeks to this loop. Accounts of the experience are mixed and some are strongly negative about hours and culture, and a few describe abrupt interviewers. That is worth weighing against the pay rather than discovering at the offer stage, and it is a fair thing to ask about directly in the manager round.

This guide covers xAI's engineering loop in depth, and the statement of exceptional work and the later rounds apply to operations, data centre and tutor roles as well. Cursor roles, in the same SpaceXAI division since August 2026, have their own track above. For management and leadership roles the loop is similar but the bar shifts to people, delivery and strategy, so pair it with the leadership interview prep hub. The bar for your exact role comes from the role-by-role guides, and the prep that actually transfers is spoken, so run a mock interview before the real one.

Also interviewing at Anthropic, OpenAI or Google DeepMind? Their guides follow the same format, from the first screen to the offer.

• 01Coditioning, xAI SWE: CV plus statement of exceptional workwhat the statement must contain, that technical team members rather than recruiters read it before the screen, that every claim is defended later, and the three reported failure modes.
• 02r/csMajors, "Worse hiring experience with xAI"an interview ended early after a disagreement rather than reaching the technical problem, a role withdrawn part-way through when the team stopped hiring for it, and a recruiter declining to give a rejection reason on company-policy grounds. Read 9 September 2026.
• 03r/xAI_community, candidate account of the full loopthe five sequential rounds and their lengths, a 30-minute recruiter screen, 30 minutes with an associate manager, a one-hour panel, a 45-minute round and a 30 to 45 minute executive interview, and the Memphis onsite that follows a successful executive round. Read 9 September 2026.
• 04r/xAI_community, application-to-offer timelinean application in late June, the first interview ten days later, a second interview six weeks after that, and an offer letter in late August. Read 9 September 2026.
• 05r/xAI_community, Grok Recruiter recorded stepthe recorded stage for multilingual tutor roles: about five questions answered one at a time, split three in English and two in the other language, with a supplied passage to translate at roughly B2 level. Read 9 September 2026.
• 06r/xAI_community, Grok Recruiter, what it gradesthat the recorded step transcribes your audio, so speaking slowly and avoiding mistakes matters more than speaking fluently at pace. Read 9 September 2026.
• 07r/xAI_community, multilingual tutor assessmentthat the round tests linguistic knowledge with practical exercises, including explaining a complex grammar point simply, and that English is the main language while fluency in the other is still checked. Read 9 September 2026.
• 08r/datacenter, xAI operations interview accountthe scenario-based technical rounds across networking, Linux, operating systems, hardware, storage, GPUs, scripting and Kubernetes, framed as real situations rather than theory, and that interviewers running late across multiple rounds was a recurring theme. Read 9 September 2026.
• 09TechPrep, xAI interview process 2026the exceptional work statement as an application filter, the 15 to 30 minute screen, the coding assessment, and the final approval stage.
• 10PracHub, xAI interview guide 2026the exceptional-work essay, the behavioural ground around ownership and acting under ambiguity, and the explicit warning that take-home details are not reliably documented and vary by team.
• 11University of Miami Toppel Career Center, get a job at xAIindependent corroboration of the application format and the reported interview stages. May 2026.
• 12Yahoo Finance, what to know about the SpaceX and xAI mergerthe all-stock merger on 2 February 2026 valuing the combined company at $1.25 trillion, and the June 2026 Nasdaq listing
• 13CNBC, Musk announces xAI re-org following co-founder departures and the SpaceX mergerthe reorganisation after the merger and the departures of the founding team
• 14CNBC, SpaceX to acquire Cursor for $60 billionthe all-stock purchase of Anysphere announced days after the listing, and Cursor's $1 billion annualised revenue
• 15Yahoo Finance, SpaceX completes the Cursor acquisitionthe deal closing on 14 August 2026 and Cursor folding into the SpaceXAI division
• 16Exponent, Cursor software engineer interview guidethe three to five rounds, the hour-long technical rounds on applied AI systems and editor primitives, and the work trial on the real codebase with AI tools allowed
• 17TechInterview, Cursor interview guide 2026the on-site work trial of up to two days on a frozen snapshot of the codebase, the review where every line is defended, and the no-product-manager, generalist bar
• 18Levels.fyi, xAI Software Engineer, United Statesthe Level 4 median, the base and stock split, the $1.2M top reported package, and the inconsistency between the two median figures. Read 9 September 2026.

Interview processes change. This reflects widely-reported and sourced conditions as of 2026 — confirm specifics with your recruiter, and treat it as a map rather than a guarantee.