Atlassian · Phone screen → Onsite
Not reported2023–2026 Atlassian Project Coding and AI Tool Use in Interviews: 11 Experiences (Part 1 :)
AI Coding finding: One candidate reports using AI on a bank app and receiving a backend part. The assigned feature and evaluation rules were not disclosed.
What the reports establish about AI
In a report posted on October 30, 2026, a candidate described coding a bank app with AI. The app appeared to have several parts, and only one was assigned. Later replies identified the backend and described uncertainty about interface changes and the boundary between implementation and design discussion.
Other candidates received different instructions. The October frontend screen required every AI feature to be disabled and checked the IDE settings over screen share. The September ML/AI product-engineering candidate said their own team used manual coding.
A separate applicant also reported an invitation to a round named “AI enabled Backend Coding” on September 19, 2026. That applicant did not share a completed interview or a task. The original author of that thread said they had not taken the round and that their own team used manual coding. The invitation and the bank-app interview are separate accounts.
Coverage: Senior/L5 frontend, P40/P50 full stack, ML/AI product engineering, and unspecified coding roles. Karat, technical screens, browser coding, Full Stack Craft, and onsite interviews. Interview periods: Q3 2023–Q4 2026. Report dates: December 2024–October 2026.
What came up
Across all eleven experiences in this two-part series, five received browser-building tasks and five discussed browser-to-server flows or login. One described using AI on a bank-app backend part.
| Topic | Reports |
|---|---|
| Building or reproducing a browser interface | 5 of 11 |
| Browser-to-server flows or login internals | 5 of 11 |
| A separate system-design question | 5 of 11 |
| An explicit AI restriction, manual-only statement, or tool incident | 4 of 11 |
| Identifiable algorithm / data-structure tasks | 3 of 11 |
| AI mentioned for mock practice | 2 of 11 |
| Behavioral / leadership rounds | 2 of 11 |
| Object-oriented player design | 1 of 11 |
| AI-assisted work on a bank-app backend part | 1 of 11 |
Candidates can appear in several rows. The AI-enabled invitation is outside the eleven firsthand interview accounts.
Questions we found
Frequencies count this article's candidates, not stored date entries. “Report” marks a posting date. Linked FastPrep records store those report dates unless another stored date is shown.
AI Coding
| Question | Frequency | All last seen dates |
|---|---|---|
| Bank-app backend part; the assigned feature was not disclosed | 1 report | 2026 Q3; report date: 2026-10-30 |
Project Coding
| Question | Frequency | All last seen dates |
|---|---|---|
| Build a Feed Grid from an API: static view, then API data | 2 reports | 2026 Q4 (report: 2026-10-12); 2026 Q3 (report: 2026-09-24) |
| Reproduce a supplied layout in plain HTML/CSS; the reference was not shared | 1 report | 2025-10-23 |
| 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 |
|---|---|---|
| Find robots whose available parts contain every requested part type | 1 report | 2025 Q4 (report: 2025-11-06) |
| Delivery Origins and Reachable Destinations | 1 report | Report: 2025-11-06, interview period 2025 Q4; additional Atlassian FastPrep history: 2025-04-20 |
| Popular Content: increase/decrease popularity and return the most popular item, with an O(1) follow-up and earliest-increased tie rule | 1 report | 2025 Q4 (report: 2025-11-06) |
| 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 |
The two ratings rows retain their different follow-ups. The interval variants in entry 07 were named only in shorthand, so they do not supply a reconstructed question here.
Low-Level Design / OOD
| Question | Frequency | All last seen dates |
|---|---|---|
| Music-player OOD: playback, queue, loop, shuffle; related Music Player Operation Sequence practice | 1 report | 2026 Q3 (report: 2026-09-19) |
System Design
| Question | Frequency | All last seen dates |
|---|---|---|
| Design a Responsive Product Catalog Search Frontend | 1 report | 2026 Q4 (report: 2026-10-12) |
| Design a Searchable Emoji Picker | 1 report | 2026 Q3; report: 2026-09-24. FastPrep stores 2026-09-23 for this same source. |
| Social Post Comments, including notifications and scaling | 1 report | 2026 Q3 (report: 2026-09-19) |
| Top-N Confluence posts | 1 report | 2025 Q4 (report: 2025-11-06) |
| 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 | 2 reports | 2025 Q4 (report: 2025-11-06); 2025 Q3 (report: 2025-09-09) |
| Explain what happens after a frontend form is submitted | 1 report | 2025 Q4 (report: 2025-11-06) |
| Trace an avatar upload through display, including different resolutions and optimization | 1 report | 2025-10-23 |
| 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 |
6 detailed candidate timelines, loops, and screens
01: L5 frontend — a six-cell feed and an explicit AI ban
Interview period: Q4 2026 · Report date: October 12, 2026 · Role: Frontend, L5 · Result: Rejected
A LinkedIn recruiter sent an invitation the day they spoke. The candidate scheduled the screen a week out and felt underprepared.
Technical screen
The local IDE was allowed, but every AI feature had to be disabled. The interviewer required screen sharing and checked the settings.
The task behind Build a Feed Grid from an API used feed records with feed_text, feed_id, and createdAt. The candidate first had to render static data in six cells, arranged as two rows and three columns, then fetch feeds from a supplied API and populate the view. React hooks and their use also came up.
There were 20 minutes for coding. The candidate struggled with the grid, later pointing to grid-template-columns: repeat(3, 1fr) as the layout they wished they had used, and did not finish.
Ten-minute system-design discussion
The next task was a searchable product-catalog frontend: a search input above a product grid, an unlimited number of products, the component tree, and fast updates after typing or filtering. The unfinished coding task had left the candidate discouraged during this discussion.
02: AI coding onsite — one backend part of a bank app
Interview period: Q3 2026 · Report date: October 30, 2026 · Stage: Onsite · Result: Rejected
The candidate explicitly said they used AI to write code. The bank app appeared to have several parts, with one selected for the interview. In a reply, the candidate said they had drawn the backend part.
Implementation and interface discussion
The candidate could not tell how much had to be implemented and how much only needed an explanation. An interface issue came up. They thought the AI's proposed change was reasonable, but said there was still an overall correctness problem. They also felt that changing someone else's interface on that scale would be unusual for a normal pull request.
The interviewer discussed design, leaving the candidate unsure whether the conversation was a hint to change the code or an open discussion. A later reply repeated that uncertainty about hints and verbal explanation.
The report does not name the assigned feature, share the repository or interface contract, or identify the AI tool, time limit, tests, or grading criteria. It establishes actual AI use in one interview, while leaving the exact backend task unknown.
03: Senior frontend — build the view, then feed it mock API data
Interview period: Q3 2026 · Report date: September 24, 2026 · Role: Senior frontend
A LinkedIn recruiter arranged this screen.
Coding and design
- In about ten minutes, build a Twitter-like feed list with HTML and CSS.
- In the next fifteen minutes, render data from a mock API in that view. This is the same feed-view family as Build a Feed Grid from an API, but this report does not specify a six-cell grid.
- Spend about fifteen minutes designing a search field and emoji picker. The interviewer asked relatively few questions, and the candidate drove the discussion.
AI permission is unstated. The practice combines this account with entry 01's grid; its fixture and tests are reconstructed.
04: ML/AI product engineering — music-player objects and a manual-only loop
Interview period: Q3 2026 · Report date: September 19, 2026 · Team: Acquired BCNY team, ML/AI product engineering · Result: Passed; later reported an offer
The two-week process emphasized communication, product thinking, and culture fit. A September 30 update confirmed an offer and said the application had covered a broad level range.
Phone screen: music player
The interviewer asked how to represent a player and stub play, pause, and next, then added a Spotify-like queue, playlist looping, and shuffling. The emphasis was object-oriented programming and communication. Music Player Operation Sequence is a bounded simulator; its deterministic shuffle and serialized outputs are practice conventions.
Onsite system design
Design a comments section for social-media posts, in the style of Instagram or Facebook. The source specified fewer than one million users and assumed login, users, and posts already existed. Discussion covered APIs, database design, scaling reads and writes, notifications, and scaling server and database resources.
Project deep dive, leadership, and behavioral rounds
The candidate discussed a system they had worked on, using project diagrams and going into individual components. They called the leadership and behavioral questions standard without giving their wording.
In replies on September 19 and 21, the author confirmed manual coding for this team. The team's AI-related work did not establish AI-tool permission.
05: P40 full stack — Karat, popularity updates, and a preparation-only GPT mention
Interview period: Q4 2025 · Report date: November 6, 2025 · Role: P40 full stack · Result: Rejected after system design
Karat: twenty minutes of web flow, thirty minutes of coding
The verbal discussion followed what happens after submitting a frontend form. The candidate felt the interviewer was checking for particular terms and repeated questions when those terms were missing. Their suggestion to practice this with GPT concerned preparation; they did not say GPT was allowed during the interview.
The first coding problem supplied parts such as Rosie_wheel, Sam_case, Sam_mirror, and Rosie_case, plus requested types wheel,case. The task was to find every robot for which all requested parts were present. In that small example, Rosie qualifies and Sam lacks a wheel.
The second problem was Delivery Origins and Reachable Destinations, called Cart Routes in the report. Given downward-directed cart-route pairs, return each start location and all terminal locations it can reach. The supplied edges were B→K, C→K, E→L, F→G, J→M, E→F, C→G, A→B, A→C, G→H, and G→I. The expected memberships were A→{K,H,I}, E→{H,L,I}, and J→{M}. The source defined N as the number of input pairs; it did not require the practice version's sorted output.
Later coding: Popular Content
Implement increasePopularity(contentId), decreasePopularity(contentId), and getMostPopular(). The initial discussion allowed O(log n). Follow-ups asked for O(1), then how to return the item first increased when several items tie. The author compared the optimization to LeetCode 432.
Full Stack Craft and system design
Craft traced login through the frontend, backend, and network, then asked about login-related work, decisions, and trade-offs in prior projects. The candidate felt it was a longer version of Karat without much depth. The system-design prompt was Top-N Confluence posts. They felt that discussion had gone well but were rejected afterward; no reason was supplied.
06: Full stack — avatar-upload discussion, then an unavailable layout reference
Interview date: October 23, 2025 · Report date: November 19, 2025 · Role: Full stack · Status at posting: No update after Karat
Karat frontend discussion and implementation
First, explain the path from clicking a profile-avatar upload button to seeing the new avatar on the profile page. The interviewer also asked about different image resolutions and optimization.
Next, reproduce a finished sample's layout using plain HTML and CSS. Time ran out as the candidate finished that portion. They thought there might have been further questions, but did not receive them. The reference image and its layout details were not included in the report, so there is no verified runnable reconstruction of this particular task.
Retake choice
An email soon offered the option to book a retake within 24 hours, before the candidate knew the first result. Asked how the attempts would count, the interviewer said both reports would go to the company. The candidate remained unsure how the company would weigh them. The report gives no AI-tool instruction.
In this series
- Part 1 (current)
- Part 2
Comments
0No comments yet — be the first to share what you know.