2026 Cursor Interviews Posted After the SpaceX Acquisition Announcement: 7 Experiences
- Company
- Cursor
- Reports
- 7 interview reports
- Level
- Full-time
- Stages
- Phone screen → Onsite
Cursor interview reports
- New Grad: one full loop from November 2025, before the deal was announced: a chat with the hiring manager, a 60-minute hash tree round inside a repo, and a day of building a feature in Cursor's own code.
- Posted after the SpaceX acquisition announcement (current): seven reports from July to October 2026: six experienced candidates (five software engineers and one machine learning engineer) and one software engineer whose hiring level wasn't given.
- Role and hiring level not reported: one technical phone screen from January 2026, a two-part per-user rate limiter.
One loop changed partway through: after the acquisition news, the candidate's team stopped being a priority and an extra behavioral round was added with a lead described as a Cursor co-founder.
Coverage: Seven candidates whose reports went up between July 14 and October 28, 2026. SpaceX announced its deal for Cursor on June 16, 2026, and candidates call the new parent SpaceXAI. Six were changing jobs while employed: five software engineers and one machine learning engineer. The seventh, a software engineer, didn't give a hiring level. Six technical phone screens and one five-round video loop. None of the seven posts says new grad.
This article keeps hiring levels and roles apart:
- Experienced, machine learning engineer: 1 report, October 2026
- Experienced, software engineer: 5 reports, July to September 2026
- Hiring level not reported, software engineer: 1 report, July 2026
Experienced: machine learning engineer
What came up
The one ML engineer screen let the candidate pick coding or system design. They picked design and got an IDE settings sync system.
Questions we found
System Design
| Question | Frequency | All last seen dates |
|---|---|---|
| Design a Versioned IDE Settings Sync System | 1 report | Report date: 2026-10-28; Additional FastPrep history: 2026-07-23 |
1 detailed candidate timeline
01: Picked system design and got the IDE settings question
Report date: October 28, 2026 · Role: Machine learning engineer · Experience: PhD, changing jobs · Result: Rejected
Technical screen
- A recruiter reached out first and then set up the screen, about two weeks before the post went up.
- The candidate could choose between coding and system design, and chose design.
- The question was the IDE settings sync system, the same one software engineer report 05 below describes.
- The rejection came two days later, with no feedback.
Looking back, the candidate thought their product design experience was thin. They hadn't led the discussion of trade-offs and bottlenecks and only got to those points when the interviewer asked.
Experienced: software engineer
What came up
Filesystem hash trees took over the coding screens: three of the five reports, posted in August and September, asked for a Merkle-style tree over files and folders. Two of them added different follow-ups: diffing two trees (02) and using the tree to cut client/server updates (04). The July reports were a design screen and a rate limiter.
| Topic | Reports | Most recent report |
|---|---|---|
| Filesystem hash (Merkle) tree | 3 of 5 | Interview in early September 2026 (report date 2026-09-22) |
| System design or design follow-up: job scheduler, hash-tree sync, IDE settings | 3 of 5 | Report date 2026-08-28 |
| Project deep dive and behavioral rounds | 1 of 5 | Report date 2026-08-28 |
| Rate limiter with pluggable rules | 1 of 5 | Report date 2026-07-18 |
A candidate can appear in more than one row.
Questions we found
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
| Filesystem Hash Tree | 3 reports | Reports: early September 2026 (interview; report date 2026-09-22), 2026-08-28 (report date), 2026-08-21 (report date); Additional FastPrep history: 2026-09-18 |
| Diff Two Filesystem Hash Trees | 1 report | Report: early September 2026 (interview; report date 2026-09-22) |
| Notification rate limiter with user, team and company limits, then pluggable rules | 1 report | Report date: 2026-07-18 |
System Design
| Question | Frequency | All last seen dates |
|---|---|---|
| Design a Scheduled Job Execution System (asked as "design CI/CD") | 1 report | Report date: 2026-08-28; Additional FastPrep history: 2026-09-18 |
| Hash-Tree Client/Server Synchronization (the second part of a coding screen) | 1 report | Report date: 2026-08-21 |
| Design a Versioned IDE Settings Sync System | 1 report | Report date: 2026-07-23; Additional FastPrep history: 2026-10-28 |
5 detailed candidate timelines, loops, and screens
02: A Merkle tree plus a diff, with every test self-written
Interview date: Around early September 2026 · Report date: September 22, 2026 · Role: Software engineer · Experience: Master's, changing jobs · Result: Rejected
Technical phone screen (60 minutes)
- The task was a Merkle tree that represents a filesystem. It works like an ordinary Merkle tree, except a node can have any number of children.
- The stub had three methods:
build(path),get_hash(path, tree)anddiff(tree, tree). The first two build and read the filesystem hash tree; the third compares two trees. - The candidate had to write all the tests themselves.
The candidate also mentioned that Cursor had been folded into SpaceXAI by then and was going by that name.
03: Five rounds, a reorg, and one more behavioral chat
Report date: August 28, 2026 · Role: Software engineer · Experience: Master's, changing jobs · Result: Rejected after five rounds
A recruiter reached out, and the interviews began before the acquisition of Cursor was formally completed. The plan was three rounds and then a one-day onsite of pair working. It became five rounds, each booked only after the previous one was passed, and never reached the pair-working day.
Round 1: Coding (60 minutes)
- The well-known Merkle tree question. The candidate felt it would be very hard to finish and get running in 60 minutes without having written one before.
Round 2: System design
- It was billed as "design CI/CD" and turned into a job scheduler design.
- In a later reply, the candidate said the interviewer seemed to make the question up on the spot. CI/CD itself barely mattered; the interviewer called it just a workload. The focus was how scheduled jobs get triggered, the task queue and how work is handed out to workers, and what happens when a worker fails.
Rounds 3 and 4: Project deep dive, then hiring manager behavioral
- Both went well.
- The day after the behavioral round, news broke that SpaceXAI had formally acquired Cursor. The candidate heard nothing for more than a week.
- Then the recruiter called. The feedback was positive, but the acquisition had brought a reorg. The original team was no longer a priority, and a new priority had come up that the candidate's background also fit. A lead from the original Cursor side, described as a co-founder, wanted to talk to them as the hiring manager behavioral round.
Round 5: Behavioral with the co-founder lead
- The lead kept it short and fast, didn't explain the new priority, and went straight into a long run of behavioral questions. The candidate felt their answers weren't really being heard.
- It ended with a thumbs-up and a "great", with no time left for the candidate's own questions.
A rejection arrived two days later. The candidate felt the long, drawn-out process had wasted both sides' time.
04: A HashTree, then using it to cut client/server traffic
Report date: August 21, 2026 · Role: Software engineer · Experience: Master's, changing jobs · Result: Rejected
Technical phone screen (45 minutes)
- Google and AI were allowed for syntax lookups, but not for writing the implementation.
- Write a HashTree over a filesystem, giving each file and folder a hash. The interviewers explained what it should do and how to hash: a text file's hash comes from applying a function to its contents, and a folder's hash comes from hashing its children's hashes joined together.
- In the second part of the same screen, use the hash tree to make sending client/server updates cheaper.
The candidate rated the screen hard and the overall experience negative.
05: An IDE settings store for users and teams
Report date: July 23, 2026 · Role: Software engineer · Experience: Bachelor's, changing jobs · Result: Passed
Technical phone screen (system design)
- Design a system that stores IDE settings. It has two kinds of entities, User and Team, and a team's settings can be pushed down to its users.
- The interviewer dug into many details, including:
- Versioning: how setting versions should work, and how updates are handled.
- Edge cases a user would notice: after a client goes offline and reconnects, how settings stay in sync without lost messages, duplicate pushes or version conflicts.
Answering a reader, the candidate added that sync was needed but worked more like a push from the server to all of a user's devices. Permissions, such as who may edit team settings, weren't explored.
06: A notification rate limiter that had to take new rules
Report date: July 18, 2026 · Role: Software engineer (backend or infra, the candidate thinks) · Experience: Bachelor's, changing jobs · Result: Passed
Technical phone screen
The candidate had just started interviewing. The task was a rate limiter, and the real test was how easily it could be extended. No prompt was shared ahead of time.
- Implement
can_send_notification(self, user_id, timestamp). It returns True if the notification can go out and False if not, and an allowed send counts once. - Limits apply at three levels: user, team and company. A helper maps a user to their team_id and company_id.
- Part 1. The candidate found it simple: keep a deque of send times and prune the expired ones.
- Part 2. The limiter had to support more rules, and the function gained a
message_sizeargument:can_send_notification(self, user_id, timestamp, message_size). The interviewer threw out new rules, such as at most 4 messages of length 50 per user per minute. The point was a pluggable design, so a later rule wouldn't touch the core limiter code. The candidate exposed aregister_handler(handler)function, with each rule living in its own handler.
The post doesn't give the actual limits for the user, team and company levels.
Hiring level not reported: software engineer
What came up
This candidate's post didn't say whether they were new grad or experienced. The screen was a Merkle tree, built on a repository the interviewers supplied.
Questions we found
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
Filesystem Hash Tree (in the report, a Merkle tree on a provided repo with get_hash(path)) |
1 report | Report date: 2026-07-14; Additional FastPrep history: 2026-09-22, 2026-09-18, 2026-08-21 |
1 detailed candidate timeline
07: A Merkle tree on a repo the interviewers provided
Report date: July 14, 2026 · Role: Software engineer (hiring level not reported) · Result: Waitlisted
Technical phone screen
- Implement a Merkle tree on a repository the interviewers provided.
- The API was a
MerkleTreeclass withget_hash(path: str).
Asked later whether that one function was the whole task, the candidate said there was probably more, but they didn't get that far.
Patterns across the reports
- Experienced software engineers kept getting a filesystem hash tree. Three of the five experienced software engineer reports (02, 03 and 04), posted between August 21 and September 22, 2026, built a Merkle-style hash tree over files and folders. The follow-ups differed: diffing two trees in 02 and using the tree to trim client/server updates in 04, while 03 doesn't mention one. Reports 02 and 03 had 60 minutes; 04 had 45.
More Cursor interview reports
- New Grad
- Posted after the SpaceX acquisition announcement (current)
- Role and hiring level not reported
Comments
0No comments yet — be the first to share what you know.