Atlassian · Phone screen → Onsite
In process2023–2026 Atlassian Project Coding and AI Tool Use in Interviews: 11 Experiences (Part 2 :)
AI Coding finding: The bank-app account in Part 1 reports actual AI use. The two tool incidents below do not establish permission to use a coding assistant.
Coverage: Part 2 follows five candidates: P50 full stack, two coding screens, a full-stack onsite, and an earlier Karat screen. Interview periods run from Q3 2023 through Q3 2025. The complete series covers eleven candidate experiences.
What came up
Three of these five candidates discussed browser-to-server or login flows. Two built browser interfaces, and two described a tool incident during a coding screen.
| Topic | Reports |
|---|---|
| Browser-to-server flows or login internals | 3 of 5 |
| Building or reproducing a browser interface | 2 of 5 |
| An explicit AI restriction, manual-only statement, or tool incident | 2 of 5 |
| Identifiable algorithm / data-structure tasks | 2 of 5 |
| A separate system-design question | 1 of 5 |
| AI mentioned for mock practice | 1 of 5 |
| Behavioral / leadership rounds | 1 of 5 |
Candidates can appear in several rows. These counts describe Part 2 only.
Questions we found
Project Coding
| Question | Frequency | All last seen dates |
|---|---|---|
| Build a File Directory Tree | 1 report | 2025 Q1 (report: 2025-01-23) |
| Build a To-do List with a GET Request, following a picture | 1 report | 2023 Q3 in the header; report: 2024-12-14 |
Algorithms / Data Structures
| Question | Frequency | All last seen dates |
|---|---|---|
| Average agent ratings, followed by a time-range query | 1 report | 2025-05-28, derived from “Wednesday” in the 2025-05-30 report |
| Stock Price Fluctuation / LC 2034, then snapshots and a timestamp-based maximum-price query | 1 report | 2025-05-28, derived from the same report |
| F1 driver with the largest last-lap improvement over previous average lap time; pit-stop follow-up | 1 report | 2025-05-27, derived from “yesterday” in the 2025-05-28 report |
| Receive 1–5 agent ratings, return descending averages, then handle monthly results | 1 report | 2025-05-27, derived from the same report |
System Design
| Question | Frequency | All last seen dates |
|---|---|---|
| Image picker; detailed requirements were not disclosed | 1 report | 2025 Q3 (report: 2025-09-09) |
Web Architecture / Full Stack Craft
| Question | Frequency | All last seen dates |
|---|---|---|
| Walk through a login's frontend, backend, and network flow | 1 report | 2025 Q3 (report: 2025-09-09) |
| Explain a login page's implementation in engineering depth | 1 report | 2025 Q1 (report: 2025-01-23) |
| Explain what happens after posting a website comment | 1 report | 2023 Q3 in the header; report: 2024-12-14 |
5 detailed candidate timelines, loops, and screens
07: P50 full stack — login internals and ChatGPT/Claude mock practice
Interview period: Q3 2025 · Report date: September 9, 2025 · Role label: P50 full stack · Source cohort field: New graduate
The sequence was one coding round and one Full Stack Craft round, followed by system design, then two behavioral interviews.
Coding
The author named two variants using the phonetic shorthand 吴流 and 酒吧流, apparently referring to LeetCode 56 and 986. That identification is an interpretation; the actual variants, inputs, and follow-ups were not described.
Full Stack Craft
The discussion began with login and covered frontend/backend interaction, TLS, DNS, React, and database partitioning. The author said it resembled the supplied preparation example and tested whether the candidate understood the concepts, without going deeply into them. They recommended ChatGPT or Claude for mock practice. That is a preparation suggestion, not a statement that AI was allowed in the round.
System design and behaviorals
The design topic was an image picker. The author felt it was straightforward and did not provide requirements. Both behavioral rounds used similar topics, with a detailed topic list from the recruiter; no exact questions were shared. Early results arrived within a day or two, while the last two rounds took more than a week. The report does not give a final hiring outcome.
08: Coding screen — ratings, stock-price snapshots, and a caption-tool dispute
Interview date: May 28, 2025, derived from “Wednesday” in the May 30 report · Stage: Code design and data structures · Outcome update: Rejected
Two coding tasks
The first task calculated average agent ratings, with a follow-up adding a time range. The second was LeetCode 2034, Stock Price Fluctuation. The candidate said they made a mistake in its follow-up and ran out of time to test.
In a June 9 reply, the same author clarified that the follow-up added a snapshot during updates and asked for the maximum price for a supplied timestamp. They did not provide enough detail to establish the snapshot API or its exact historical-query semantics.
Tool incident and outcome
On July 7, the author reported rejection and said they had used Google caption, which the interviewer described as an AI tool. Neither a causal link to the rejection nor permission for a coding assistant was established. The July update is part of the May interview account, not another interview.
09: Coding screen — F1 improvement, monthly ratings, and an unexpected phone activation
Interview date: May 27, 2025, derived from “yesterday” in the May 28 report · Result: Rejected the next day
First coding round
Given drivers and their lap times, identify whose last lap improved most compared with their average over previous laps. The follow-up asked how pit stops should be handled. The author did not disclose the intended pit-stop rule or their implementation.
Second coding round
For a support-ticket system, accept customer ratings on a five-point scale, return each agent's average, and sort agents from highest to lowest average. The follow-up asked how to handle monthly results.
The candidate said ChatGPT on their iPhone unexpectedly activated during this round and began summarizing after they had written the code and explained it. They muted it, explained the unexpected activation, and received an “OK” from the interviewer. The rejection arrived the next day. The report gives neither permission to use AI nor a confirmed reason for rejection.
10: Full-stack onsite — directory-tree UI and a separate Craft discussion
Interview period: Q1 2025 · Report date: January 23, 2025 · Role: Full stack
Browser coding
Build a file directory-tree interface from a simple supplied mock and data. The candidate could use a local IDE or a tool such as CodeSandbox, debug in the browser, and choose a framework freely. The report did not provide the original mock, data, or detailed interactions; the practice exercise supplies a bounded version.
Full Stack Craft
Explain a login page's implementation in depth, with practical engineering knowledge. In a later reply, the author confirmed that this Craft discussion did not require coding. That clarification does not remove the separate browser-coding round above.
Permission to choose an IDE or framework does not establish permission to use AI, which the report does not address.
11: Earlier Karat — a reference-picture to-do list and a GET request
Interview period: Q3 2023 in the source header · Report date: December 14, 2024 · Result: Invited to onsite
An old colleague referred the candidate. HR contacted them about a month later and arranged Karat.
Verbal discussion, then browser coding
The interviewer first asked what happens after a comment is posted on a website. The coding task supplied a to-do-list picture and asked the candidate to make a JavaScript GET request and reproduce the displayed result.
Build a To-do List with a GET Request preserves that task's GET-and-render core. The source did not share the picture, endpoint, response data, or interactions, so the practice is a reconstruction rather than a recovery of its exact appearance.
Three days later, the candidate was invited onsite. AI permission is unstated. The 2024 report and 2023 header period remain separate.
Patterns across the reports
These patterns compare both parts of the series.
- The feed task had a static step and an API step. Both 2026 frontend screens, entries 01 and 03, used that sequence. Only entry 01 specified the two-by-three layout, record fields, React-hook discussion, and AI restriction. Those details should not be transferred to the other candidate's interview.
- Full Stack Craft could sit beside a separate coding task. Entries 05, 07, and 10 describe a login-centered Craft discussion as well as another coding round. Their reported depth differed: entry 10 emphasized a deep dive, while entries 05 and 07 described shallower conversations.
- Similar agent-rating tasks led to different follow-ups. The two May 2025 screens, entries 08 and 09, both involved averages. One added a time range, while the other asked about monthly results. Neither report supplies the other's missing rules.
In this series
- Part 1
- Part 2 (current)
Comments
0No comments yet — be the first to share what you know.