Cursor interview, decoded.

Updated · Sources at the end

Cursor hires every engineer and designer through a two-day onsite work trial, and its CEO has said why: a traditional coding interview cannot tell you whether someone can go end to end in a codebase, or what they would build if left alone. Before the trial come a recruiter or manager screen that checks you actually use the product, and one to three technical rounds of about an hour where AI tools are banned beyond autocomplete (Michael Truell: programming without AI is still a great time-boxed test of skill) and the problems lean toward editor primitives rather than puzzles: a syntax-aware edit, streaming model output into a buffer, a file-tree diff, a context retrieval system, a Merkle tree to detect changes between repository snapshots. Then the trial: a desk, a laptop, a frozen snapshot of the real code, meals with the team, and at the end a demonstration of what you built and a review where every line is defended. The head of talent calls it expensive and worth it, and says most companies get it wrong by trying to get work out of the candidate.

Anysphere, Cursor's maker, was bought by SpaceX in an all-stock deal announced in June 2026 and closed on 14 August, and now sits in the SpaceXAI division alongside xAI; a posting may carry any of the three names, and the loop described here is Cursor's own, distinct from Grok's. Levels.fyi puts the US median for a Cursor software engineer at about $1.04 million, the highest on the site, read 22 September 2026.

Free · 2 minutes · no account

The questions for your exact Cursor role and level.

See the questions Cursor asks ↓Run a free mock interview →

Interviewing at Cursor? Below are the questions candidates report. For the ones your exact role and level will get, paste the posting. Your first mock is free.

01

Who Cursor hires

Cursor hires software engineers across the editor, models, infrastructure and product, plus designers who go through the same trial, in San Francisco and New York, on site. Anysphere ran with no product managers and few management layers, so the bar is a generalist who owns a feature end to end, and every account says the same thing about the first screen: you have to be a real user of the product and able to say why this product rather than a hot company. Comfort with AI coding tools is expected in the trial; the screens test what you can do without them.

02

What is the Cursor interview process?

The CEO and the head of talent have described the shape on record, and the two 2026 guides agree on the details below. Published accounts date from before and after the acquisition, so expect drift in names and timing rather than in the trial itself.

01
Recruiter or manager screen
30 to 45 minutes

Genuine interest in the product, evidenced by using it; what you have shipped end to end; why Cursor rather than a hot company. Accounts say a candidate who has not used the editor does not pass this call.

02
Technical rounds
One to three rounds of about 60 minutes, no AI tools beyond autocomplete

Applied problems on the product's own primitives: a syntax-aware edit operation, applying multi-file model edits while the user keeps typing, streaming LLM output, a file-tree diff, a text buffer with efficient insert, delete and line lookup, a context retrieval system, a Merkle tree over repository snapshots. Truell has said the ban on AI is deliberate: a time-boxed test of skill, and fair to strong programmers who have not used the tools.

03
Take-home
Asynchronous, similar weight to an onsite round

Reported for some senior roles in place of a screen.

Some senior roles
04
Onsite work trial
Up to two days on site with the team

A frozen snapshot of the real codebase, a laptop, and in Truell's framing no assigned task: what would you build if left in a vacuum. You scope it, build it with the team's tools, and demonstrate it at the end; the review goes line by line. Meals with the team are part of the assessment in both directions.

05
Founder and team conversations
Around the trial

Fit both ways; founder-level conversations for many roles. Truell has said the two days exist partly to find people passionate about the problem space rather than shopping for a job.

03

Cursor's two-day work trial: what it is and how to pass it

Reported problems and how to solve them

The trial is the decision round, and it is described on record: every engineer and designer does it, it runs up to two days on site, and the material is a frozen snapshot of the real codebase. Its intent is stated too, which is the useful part; you are not guessing what is graded. Here is what the two days ask and how the people who clear them use the time.

  • Day one, first two hours: read before you build: Orient in the snapshot: how a feature is wired from the editor surface to the model call, where tests live, how the team runs things locally. Ask the engineers around you; the trial is with the team, and how you ask is observed. Candidates who start coding in the first half hour report building the wrong thing well.
  • Scope something you can finish: The task is yours to pick, and Truell's framing is what you would build if left alone. Choose a real, small improvement to the product you use: something you have wanted as a user, reachable in a day and a half, with a demo at the end. Ambition is read from the choice, completion from the demo.
  • Build with the team's tools, tested: The trial uses the tools engineers use day to day; the AI ban is for the screens. Write it the way the codebase is written, with the tests the codebase expects, in small commits you can walk through. The review at the end goes line by line, so every line needs a reason.
  • The demo and the review: Show it working in the product first, then the design, then what you would do next. Expect the review to press on choices; defend them or concede them cleanly. A candidate who says which part is a hack and why scores better than one who hides it.
  • Meals and conversations count: The trial is a cultural interview both ways, in the CEO's words. Ask what the team is working on, what the product should do next, how decisions get made without product managers. Truell has said the two days exist to find people who care about the problem; that shows in the conversations more than in the code.
  • Ask about terms before you go: Public accounts differ on whether the trial is paid and how IP is handled. Ask the recruiter directly, in writing, what compensation and IP assignment apply to the two days, before you fly.

Prepare by doing the trial to yourself: pick an open-source editor or a large TypeScript codebase you have never read, give yourself a day, and ship a small feature with tests and a demo. Then practise the screens without AI: a text buffer, a diff, a Merkle tree over a directory, streaming output into a buffer, each in under an hour.

The rest of the loop depends on your role. Pick it, or paste the posting, and Calibrd predicts the rounds and questions you are likely to get at Cursor.

Predict my Cursor questions →
04

What Cursor screens for

Cursor publishes no values list; its founders and its head of talent have said what the trial is for.

  • Can you go end to end in a codebase you have never seen: the trial's first question, in the CEO's words
  • What would you build if left alone: the second
  • Genuine use of the product, checked at the screen
  • Bottom-up experimentation and independent team-building, which Truell says the company prioritises
  • Programming skill without AI assistance, tested deliberately
05

Cursor interview questions

Candidate-reported themes

The reported problems are the product's own internals; the trial's question is the one the CEO has stated publicly.

Behavioural & motivation

  • Why Cursor, and how do you use it today?Listening for: Specificity · Connection · Research
  • You have two days in our codebase and no assigned task. What do you build?Listening for: A real, specific feature · Scoped to finish · How you would start
  • Tell me about something you took from idea to shipped on your own.Listening for: How they noticed · Why it mattered · How they took it on
  • Why not put the whole repository into a long context window?Listening for: Cost and latency · Answer quality · When retrieval wins

Technical

  • Design a text buffer with efficient insert, delete and line lookup. Which data structure would you use, and why?Editor primitives: A text buffer with efficient insert, delete and line lookup; a syntax-aware edit operation; a file-tree diffListening for: Picks a fitting structure · Fast line lookup · Costs stated
  • How would you apply multi-file edits streamed from a model while the user keeps typing in the same files?Applying model output: Streaming LLM output into a buffer; applying multi-file edits while the user keeps typing; fast, reliable application of a 500-line rewriteListening for: Anchors edits reliably · Handles overlapping edits · Streams without lag
  • How would you index a 100k-file monorepo for AI-assisted retrieval, with live updates as files change?Retrieval over code: Index a 100k-file monorepo for AI-assisted retrieval with live updates; a context retrieval system; why not stuff the repo into long contextListening for: Chunks code sensibly · Keeps the index fresh · Retrieval and ranking
  • Use a Merkle tree to detect which files changed between two snapshots of a repository. How does it work, and why is it fast?Change detection: A Merkle tree to detect file changes between repository snapshotsListening for: Builds the tree correctly · Prunes unchanged subtrees · Edge cases
  • How would you serve Tab next-edit prediction under 100 milliseconds for millions of daily users?Serving and evaluation: Tab next-edit prediction under 100 milliseconds; offline and online evaluation for code-editing models; an inference-cost model for millions of daily usersListening for: Latency budget broken down · Cost at scale · How you would measure it
  • Design a harness that lets an agent run multi-file tasks, iterating on builds, tests and lints without disturbing the user.Agents: A harness for multi-file agent tasks with guardrails; agents iterating on builds, tests and lints without disturbing the userListening for: Isolated from the user · Feedback loop and stops · Guardrails and review

In Cursor's work trial you build with AI tools and then defend every line in the review. For a quick warm-up on catching AI mistakes, try spotting the bugs in short pieces of AI-written code. Spot the AI bug, about 35 seconds →

These are the reported themes, and your loop is role-specific. Paste the actual posting and Calibrd predicts the questions for that exact role and level. How the questions are predicted →

Predict my Cursor questions →
06

Pay

Software engineer, US median~$1.04M total
Software engineer, San Francisco Bay Area median~$1.2M total

Levels.fyi medians for Cursor software engineers, updated 17 September 2026 and read on 22 September, from a small sample that skews senior; reported packages run from under $200K to well over $1 million, most of it equity. Since 14 August 2026 that equity is SpaceX stock through the acquisition; ask what the grant converts to, on what schedule, and how the SpaceX listing affects liquidity. Third-party figures of $808K to $1.28 million are consistent with the Levels.fyi range.

07

How to prepare for a Cursor interview

  1. Use the product for real before applying, and have opinions about it. The first screen checks this, and the trial's open task rewards it.
  2. Practise the screens without AI assistance: a text buffer with insert, delete and line lookup; a file-tree diff; a Merkle tree over a directory; streaming output applied to a buffer. Under an hour each.
  3. Run a self-trial: a codebase you have never read, one day, a small feature with tests and a demo.
  4. Prepare your answer to the stated question: two days in the codebase and no assigned task. Have two candidate features in mind, one safe and one ambitious.
  5. Be ready to explain why long context does not replace retrieval, and how you would evaluate a code-editing model offline and online.
  6. Ask about trial pay and IP in writing before the onsite.

This guide covers Cursor's engineering and design loop, which is the same for both and distinct from the Grok loop described in the xAI guide, though both now sit in SpaceX's SpaceXAI division. 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 ElevenLabs, Sierra or Scale AI? Their guides follow the same format, from the first screen to the offer.

Knowing the questions isn’t the same as answering them out loud. Run a Cursor mock: spoken answers, coached on the spot. Your first mock is free.

Run a Cursor mock →
08

FAQ & sources

What is Cursor's interview process?

Published accounts date from before and after the acquisition, so expect drift in names and timing rather than in the trial itself. The stages, in order: Recruiter or manager screen; Technical rounds; Take-home; Onsite work trial; Founder and team conversations.

What does Cursor look for in candidates?

Cursor publishes no values list; its founders and its head of talent have said what the trial is for. Can you go end to end in a codebase you have never seen; What would you build if left alone; Genuine use of the product, checked at the screen; Bottom-up experimentation and independent team-building, which Truell says the company prioritises.

What questions does Cursor ask in interviews?

The reported problems are the product's own internals; the trial's question is the one the CEO has stated publicly. Technical themes: Editor primitives; Applying model output; Retrieval over code; Change detection; Serving and evaluation; Agents.

What is the Cursor two-day work trial?

The trial is the decision round, and it is described on record: every engineer and designer does it, it runs up to two days on site, and the material is a frozen snapshot of the real codebase. Its intent is stated too, which is the useful part; you are not guessing what is graded. Covered on this page: Day one, first two hours: read before you build; Scope something you can finish; Build with the team's tools, tested; The demo and the review; Meals and conversations count; Ask about terms before you go.

How do I prepare for a Cursor interview?

Use the product for real before applying, and have opinions about it. The first screen checks this, and the trial's open task rewards it. Practise the screens without AI assistance: a text buffer with insert, delete and line lookup; a file-tree diff; a Merkle tree over a directory; streaming output applied to a buffer. Under an hour each.

Prep for a real Cursor role

Practise your Cursor interview, out loud.

Paste a real Cursor posting and Calibrd predicts the questions for that role and level, benchmarks the pay, and flags the gaps an interviewer will probe in your CV — then listens to your spoken answers and coaches them. Your first mock is free.

Free to start · No card · Encrypted at rest, never used to train AI, remove anytime

Two weeks out?

Email me the Cursor prep checklist

The round that decides it, the questions they ask, and what to do before, in one email. Nothing else unless you sign up.

Cursor Interview: Two-Day Work Trial, Screens, Pay — Calibrd