2024–2026 Apple AI Coding & Related Interviews: 17 Experiences, New Grad First
- Company
- Apple
- Reports
- 17 interview reports
- Level
- New Grad
- Stages
- Phone screen → Onsite
A Beijing MLE New Grad screen ended with about 15 minutes on AI coding habits, model differences, and prompts. Two other New Grad reports describe LLM SDK, RAG, TF-IDF, and generator tasks.
Coverage: 17 independent Apple applications: three explicitly New Grad, plus internship, experienced-hire, and ICT4 accounts. Known interview periods span Q4 2024 to Q3 2026. New Grad comes first.
What came up
Three candidates explicitly reported AI permission or access during coding. Others describe verbal discussion, a pending round, or an advertised AI round that became ordinary coding. The accompanying ML and project rounds disclose more complete tasks.
| Topic | Reports |
|---|---|
| AI-round plans, tool use, or AI work habits | 9 of 17 |
| ML/GenAI implementation, knowledge, or design | 9 of 17 |
| System or code design | 7 of 17 |
| Modifying or debugging existing code | 3 of 17 |
| Explicit permission or access to AI during coding | 3 of 17 |
| AI in the round name or advance notice, but no actual AI use | 2 of 17 |
Topics overlap. Question counts reflect this article; extra Apple dates do not add reports.
Questions we found
Dates distinguish interviews, original reports, and extra Apple practice history. Frequencies count reports.
AI-assisted coding
| Disclosed task | Frequency | All last seen dates |
|---|---|---|
| Class usage example with AI; class/interface undisclosed | 1 | Interview: Q3 2026; report: 2026-09-23 |
| Complete a function with search and AI; task undisclosed | 1 | Report: 2026-04-02 UTC / 04-01 locally; interview described as “recently” |
Verbal AI-workflow questions
| Question | Frequency | All last seen dates |
|---|---|---|
| How do you use AI to code, how do models differ in practice, and which prompts improve generated code? | 1 | Interview: Q4 2025; report: 2025-12-27 |
| If AI were available, how would you use it for the coding task? | 1 | Interview: Q2 2026; report: 2026-06-10 |
| How do you like to use AI, and does it actually improve productivity? | 1 | Interview: Q4 2025; report: 2025-11-06 |
Algorithms / data structures
| Question | Frequency | All last seen dates |
|---|---|---|
| A 3Sum-like task, with two-pointer optimization and complexity follow-ups | 1 | Interview: Q4 2025; report: 2025-12-27; extra Apple history: 2020-11-30 |
| A Time-Based Key-Value Store variation, followed by thread safety | 1 | Report: 2026-09-05 |
| A substring-match generator that yields True on a match | 1 | Interview: Q3 2025; report: 2025-11-21 |
| Array deduplication; constraints undisclosed | 1 | Interview: Q4 2025; report: 2025-11-06 |
| Deck shuffling/order handling; exact rules undisclosed | 1 | Interview: Q3 2025; report: 2025-08-14 |
ML algorithm implementation
| Question | Frequency | All last seen dates |
|---|---|---|
| Simple TF-IDF | 1 | Interview: Q3 2025; report: 2025-11-21 |
| Implement beam search | 1 | Interview: Q3 2025; report: 2025-09-13 |
| Implement attention; this report does not specify head count | 1 | Interview: Q3 2025; report: 2025-09-13; extra Apple history: 2025-08-15 |
| Implement KNN from scratch | 1 | Interview: Q4 2024; report: 2025-05-21 |
| Implement circular padding | 1 | Interview: Q4 2024; report: 2025-05-21 |
Classes / interfaces
| Question | Frequency | All last seen dates |
|---|---|---|
| LLM SDK output classes: emit hello, or append world to the input | 1 | Interview: Q4 2025; report: 2025-10-15 |
Project coding / debugging
| Question | Frequency | All last seen dates |
|---|---|---|
| Debug a PyTorch UNet | 1 | Interview: Q4 2025; report: 2026-02-25 |
| Complete a DAG task pipeline: add tasks and calculate results | 1 | Interview: Q4 2025; report: 2025-10-07 |
| Build a model-inference API | 1 | Interview: Q4 2025; report: 2025-10-07 |
| Extend a three-bot rock-paper-scissors game: configurable rounds and draw counts | 1 | Interview: Q3 2025; report: 2025-08-14 |
Concurrent programming
| Question | Frequency | All last seen dates |
|---|---|---|
| Manually create threads to upload images concurrently | 1 | Interview: Q4 2025; report: 2025-11-06 |
System design
| Question | Frequency | All last seen dates |
|---|---|---|
| Ticket booking; requirements and scale undisclosed in this account | 1 | Report: 2026-09-05; same account recorded in practice history: 2026-09-06 |
| Chatbot ML system design | 1 | Interview: Q2 2026; report: 2026-06-10 |
| RAG with multiple input modalities | 1 | Interview: Q3 2025; report: 2025-11-21 |
| Chat/registration systems; boundaries undisclosed | 1 | Interview: Q4 2025; report: 2025-11-06 |
| Local-business reviews: read/write scaling, caching, and CDN | 1 | Interview: Q3 2025; report: 2025-08-14 |
| Short links | 1 | Interview: Q3 2025; report: 2025-08-14 |
Low-level design / code review
| Question | Frequency | All last seen dates |
|---|---|---|
| Review code and suggest an improving design pattern | 1 | Report: 2026-09-05; code, domain, and pattern undisclosed |
ML / LLM evaluation
| Question | Frequency | All last seen dates |
|---|---|---|
| Evaluate grounded assistant responses: tool results, hallucinations, evaluation, and long context | 1 | Interview: Q1 2026; report: 2026-03-27 |
Technical knowledge
| Question | Frequency | All last seen dates |
|---|---|---|
| Spring versus Spring Boot; how Spring Boot starts | 1 | Report: 2026-09-05 |
| ConcurrentHashMap and how threads work | 1 | Report: 2026-09-05 |
| Dataset size, overfitting/underfitting, and model development | 1 | Interview: Q4 2025; report: 2026-02-25 |
| Super-resolution MSE versus perceptual loss; UNet and skip connections | 1 | Interview: Q4 2025; report: 2026-02-25 |
Behavioral / project experience
| Question | Frequency | All last seen dates |
|---|---|---|
| Favorite project | 1 | Dates unavailable; account 17 |
| Recently encountered bugs | 1 | Dates unavailable; account 17 |
| Whether the candidate has AI project experience | 1 | Interview: Q3 2025; report: 2025-08-14 |
New Grad: three accounts
01: New Grad MLE — AI coding habits
Report: 2025-12-27 · Interview: Q4 2025 · PhD / New Grad · Beijing · Rejected
The screen covered recent projects and current time-series models; about 20 minutes on an easy 3Sum-like task, with two-pointer optimization and complexity follow-ups; then about 15 minutes on AI coding.
The AI questions were specific:
- How do you use AI to write code?
- How do different models behave differently in your actual use?
- Which prompts improve the generated code?
The editor link arrived at interview time. AI permission for the algorithm task and its full prompt were not reported.
02: iOS GenAI Health — LLM SDK output classes
Report: 2025-10-15 · Interview: Q4 2025 · Master's / New Grad · Technical screen
The task was an LLM SDK output base class and subclasses: one emits hello; another emits the input followed by world.
The LLM already ran elsewhere; no network request, URL session, or exception wrapper was needed. The interviewer wanted the output function outside the session. Unsure about that boundary, the candidate neither finished nor ran the code.
The practice version adds an explicit interface and output format. AI-assistant permission and the final outcome were not reported.
03: New Grad MLE — RAG, TF-IDF, generator
Report: 2025-11-21 · Interview: Q3 2025 · PhD / New Grad · Rejected the next day
The team supported hardware using RAG. After the resume, the HM spent about 20 minutes on RAG accepting multiple input modalities; the candidate drew the design.
Two technical screens included technology-selection questions, simple TF-IDF pseudocode, and a substring-match generator that yields True when a substring matches a pattern.
The practice adds overlap, empty-pattern, and collected-result rules absent from the account. AI-assistant permission was not reported.
Other candidate accounts
04: Software intern — class usage with AI
Report: 2026-09-23 · Interview: Q3 2026 · Software internship, titled 2027
Three days after applying online, the candidate heard back. Introductions and a short resume discussion preceded a custom class-usage task with explicit AI permission.
Class purpose, interface, inputs/outputs, model, and duration were undisclosed. The editor lagged. The candidate struggled with C++ questions; the interviewer allowed search and gave positive feedback. The candidate expected rejection, without a confirmed outcome. The 2027 title does not establish New Grad status.
05: ICT4 — advertised AI, actual ordinary coding
Report: 2026-09-05 · Update: 2026-09-21 · Java / Spring Boot · ICT4 · Nine years' experience
The recruiter announced AI-assisted coding, design, and leadership-style rounds. On September 21 the candidate twice clarified that actual coding had no AI assistance.
Earlier rounds included:
- A slight Time-Based Key-Value Store variation, then thread safety using
ConcurrentHashMap. The variation and concurrent contract were undisclosed. - Identify a design pattern to improve supplied code; code and pattern undisclosed.
- Ticket-booking design. Seat, payment, and queue details in the practice version were not disclosed by this candidate.
- Spring versus Spring Boot, its startup process, ConcurrentHashMap, and threads.
A promised later write-up was not linked. The announced AI format changed before the actual interview.
06: ICT4 — AI pairing interrupted
Report / signed offer: 2026-07-28 · Services business team · ICT4 · Six years' experience
- June 24: a 45-minute phone screen.
- July 9 and 13: virtual Onsite, later confirmed as two coding and three design sessions.
- July 28: signed the official offer.
One session offered AI pairing through a right-sidebar agent selector. It disconnected after about 20 minutes and could not reconnect; the interviewer accepted verbal explanation afterward. Onsite slots were described as 45 minutes, but this session's actual finish time was not given.
The candidate thought it assessed directing AI, explaining requirements/architecture constraints, and checking robustness. Their advice emphasized design patterns, tests, edge cases, and treating AI as a junior collaborator. This was their interpretation, not an official rubric.
Exercise, model, acceptance criteria, and task count were undisclosed; a question about three tasks went unanswered. Java/Python were probably acceptable in the candidate's view, subject to team/job requirements.
07: Multimodal MLE — verbal AI coding
Report: 2026-06-10 · Interview: Q2 2026 · Master's / experienced hire · Rejected after a wait
The phone screen covered Transformer-related ML coding and knowledge. Onsite comprised regression coding, chatbot ML system design, ML/generative-model fundamentals, then a session labeled AI coding.
No AI tool was available in the last session. The candidate briefly explained how they would use AI; running the program was unnecessary. Exercise, rubric, and duration were undisclosed. A reader's one-hour suggestion was unconfirmed. The rejection's cause was not given.
08: A late added AI round — notice only
Report: 2026-04-24 · Experienced SDE applicant · Positive senior-manager feedback
The candidate expected a higher-level director conversation. The recruiter instead said an AI-assisted coding round would be added from April 27, using AI in an online editor. Details were unavailable; language choice would depend on the interviewer.
The candidate had not taken this new round when posting. A follow-up from the same application placed the anticipated manager/director stages around rounds seven and eight. No exercise, model, duration, or result was disclosed. The recruiter's broad wording does not establish the same process across Apple teams.
09: Complete a function with search and AI
Report: 2026-04-02 UTC / 04-01 locally · Interview described as recent
The candidate described their own Apple interview: technical career-background questions, followed by completing a function in an online coding tool. Search and AI were allowed, and the interviewers were friendly and communicative.
Function, language, model, duration, outcome, level, and location were undisclosed. An unexplained job abbreviation leaves the exact role and New Grad status unknown.
10: Siri MLE — evaluation instead of coding
Report: 2026-03-27 · Interview: Q1 2026 · Master's / experienced hire
A 30-minute HM discussion covered projects and background; the candidate heard they passed a week later. The next session was advertised as 60 minutes of ML coding, but involved essentially no coding.
Instead, it examined generated assistant responses and their grounding: how to evaluate answers, ensure support from tool results, explain and mitigate hallucinations, run evaluation, and manage excessive conversation history or user-context length. AI-assistant permission and a final hiring outcome were not reported.
11: Image/video MLE — PyTorch UNet debugging
Report: 2026-02-25 · Interview: Q4 2025 · Master's / experienced hire · Rejected
Questions covered dataset size, overfitting/underfitting, model development, super-resolution MSE versus perceptual loss, UNet, and skip connections. Coding involved debugging PyTorch UNet code.
The candidate tentatively recalled input tensor shape, a decoder boolean, final kernel size, and num_classes. They also emphasized the original-paper 572 input / 2 × 388 × 388 output and discussed version-dependent shapes in replies. These were recalled changes and preparation advice; original code/tests and a verified defect list were absent.
The UNet debugging exercise uses real PyTorch with reconstructed code/tests. Its exact faults and acceptance checks are practice additions. AI-assistant permission was not reported.
12: Senior Onsite — AI productivity questions
Report: 2025-11-06 · Interview: Q4 2025 · Senior / experienced hire · Rejected a week later
Ordinary technical rounds mentioned isolated integers, ticket itineraries, concurrent image uploads using manually created threads, array deduplication, and chat/registration design. Full task rules and constraints were not supplied.
The HM asked how the candidate liked to use AI and whether it improved productivity. Their answer was undisclosed. This was HM discussion; a separate AI practical round or AI permission for earlier tasks was not reported.
13: AIML Infra — DAG, debugging, inference API
Report: 2025-10-07 · Interview: Q4 2025 · Bachelor's / experienced hire · Rejected
The screen involved JSON cleaning/processing. Onsite included:
- Complete an existing DAG task pipeline, adding tasks and calculating results.
- Repair file/data-processing code; the script and defects were not disclosed.
- Projects and behavioral discussion.
- Expose model inference through a Web API, using Flask.
- An HM interview.
AI-assisted format or tool permission was not reported for these engineering tasks.
14: ML research — implement beam search and attention
Report: 2025-09-13 · Interview: Q3 2025 · PhD / experienced hire · Rejected
The account describes six technical sessions, a research talk, and behavioral discussion. ML coding included beam search and attention, plus an unnamed ordinary coding exercise.
The attention practice uses multiple heads; this report does not specify head count, masks, or projection requirements. Research/knowledge topics included reasoning, decoding acceleration, window/paged attention, quantization, and training. No AI coding-assistant workflow was reported.
15: Modify a three-bot rock-paper-scissors game
Report: 2025-08-14 · Interview: Q3 2025 · Master's / experienced hire
Coding involved deck shuffling/order handling, then DFS, and modifying an existing three-bot rock-paper-scissors game: make the number of rounds configurable and count draws.
Design covered a local-business-review service, including read/write scaling, caching, and CDN, plus short links. Two HM rounds discussed AI project experience, learning, feedback, and projects. The author used AI to organize the written account; that does not establish interview tool permission. The final hiring outcome was not confirmed.
16: CVML — KNN and circular padding
Report: 2025-05-21 · Interview: Q4 2024 · PhD / experienced hire · Rejected
The screen required implementing KNN from scratch. Onsite covered ordinary algorithm variants, Transformer/ViT/diffusion, generated-image evaluation, ML fundamentals, and implementing circular padding.
The interview was in late 2024, before its 2025 report. These ML tasks had no reported AI-assistance workflow.
17: ICT4 — two AI Programming parts
ICT4 · Eight years of experience · Exact interview date unavailable
The candidate included Apple in a year of applications. The recruiter discussed the company/team product. They solved an unnamed easy coding task and follow-ups. A roughly 45-minute HM covered background, a favorite project, and recent bugs.
The Onsite header said six rounds, but the narrative listed five:
- A 30-minute cross-functional conversation with a nontechnical interviewer; questions undisclosed.
- A second HM discussion.
- Coding with verbal multithreading follow-ups.
- System design; system identity undisclosed.
- AI Programming: the candidate finished two parts but did not understand the evaluation criteria.
AI task, tool, model, duration, and acceptance criteria were undisclosed. Two parts do not establish two questions. The author reported rejection across the listed companies, without Apple's specific rejection stage or feedback. The six-round/five-session mismatch remains unresolved.
AI coding rounds at other companies
- Meta: Meta AI coding interview: two onsite rounds where you finish a hard algorithm inside a mostly written codebase with an AI model, then face stress-test suites.
- Google: Google AI coding interview: an online assessment with HackerRank's built-in assistant, and a live round on a multi-file project where using AI was optional.
- Stripe: Stripe AI coding interview: experienced engineers kept getting the same transaction rule parser, built step by step with a built-in model.
- Stripe interns: Stripe intern AI coding round: ten intern accounts of the same AI-assisted round.
- Rippling: Rippling AI coding interview: HackerRank's built-in AI on Rippling's familiar problems, plus an HR codebase at the onsite.
- DoorDash: DoorDash AI coding interview: AI CodeCraft: build and run several small services live with your own AI tool.
- LinkedIn: LinkedIn AI coding interview: one long, multi-part task in an IDE whose agent edits files directly.
- Shopify: Shopify AI coding interview: pair programming with your own AI on Shopify's well-known question set.
- Datadog: Datadog AI coding interview: bring any AI tool and share your screen; the tasks are easy, and the follow-ups decide the round.
- Capital One: Capital One AI coding interview: new grad Power Day agentic round: fix a bug, then add a feature to a large full-stack app with Claude Code.
- Bloomberg: Bloomberg AI coding interview: intern rounds that start in a small file and move to a large codebase, debugging with AI help.
- xAI: xAI AI coding interview: new grad take-homes and onsites where you build with AI.
- Adobe: Adobe AI coding interview: AI showed up as one step in a director round and in a take-home; the rest stayed classic.
- Snowflake: Snowflake AI coding interview: design and build a rate limiter or a retrying task queue while the AI writes the code; one AI Engineer onsite was in person with your own laptop.
- Snap: Snap AI coding interview: a backend-only chat service from an empty Python project with Claude or Codex, or a feature in an existing codebase.
- Ramp: Ramp AI coding interview: build inside a repo Ramp provides; one repo logged every prompt for the interviewer to read.
- Anthropic: Anthropic AI coding interview: review four pull requests, then improve one with Claude Code.
More companies with AI rounds on FastPrep:
- Amazon: Amazon interview report: an AI-assisted debugging OA (1 experience).
- Atlassian: Atlassian interview report: project coding and AI tool use (11 experiences, two parts).
- Circle: Circle interview report: three AI rounds: building, debugging and PR review (1 experience).
- Volta: Volta interview report: a deliberately vague data task with a token budget (1 experience).
Across companies: 2026 AI Coding OAs at 15 Companies: 32 Reported Problems and 2025–2026 AI Coding Phone Screens and Onsites at 43 Companies: 76 Reported Problems.
Comments
0No comments yet — be the first to share what you know.