2025–2026 AI Coding Phone Screens and Onsites at 43 Companies: 76 Reported Problems
An OA lets you work alone. A phone screen or onsite puts an interviewer next to you while you work in a real repo, and that changes what gets asked. This article covers the 76 AI coding problems FastPrep collected from live rounds at 43 companies in 2025 and 2026.
The short version: live rounds ask you to build far more than they ask you to fix. 55 of the 76 problems are build or extend tasks, and only 17 are bug fixes, the opposite of the OAs.
Coverage: Project-style AI coding tasks seen in 2025 and 2026 in phone screens (29 problems) and onsites (53) at 43 companies, plus one problem whose company wasn't named. Six problems were reported in both rounds. Levels run from junior (16) through mid level (52) to senior (8). Only 29 of the 76 problems say whether the candidate was an intern, new grad, or experienced hire. Sightings run from 2025-01-10 to 2026-10-31.
What came up
Almost every live-round task is graded by tests, and most of them ask you to shape an API and a data model while someone watches. Debugging matters less here than in OAs, showing up in 22 of the 76 problems.
| Topic | Problems | Most recent sighting |
|---|---|---|
| Testing | 62 of 76 | 2026-10-31 |
| Data modeling | 51 of 76 | 2026-10-31 |
| API design | 50 of 76 | 2026-10-31 |
| Debugging | 22 of 76 | 2026-10-31 |
| Concurrency | 14 of 76 | 2026-10-29 |
| Security | 2 of 76 | 2026-10-22 |
| Transactions | 7 of 76 | 2026-10-06 |
A problem can count toward several topics, so the rows add up to more than 76.
Across the 76 live-round problems, 28 ask you to complete or extend existing code, 27 are build-from-scratch, 17 are bug fixes, 3 are ML pipelines, and 1 is a code review. Next to the OAs from the same two years, the shift is clear: bug fixes are 19 of 32 OA problems, but 13 of 53 onsite problems.
| Round | Bug fix | Complete or extend | Build from scratch | ML pipeline | Tests or review | Total |
|---|---|---|---|---|---|---|
| OA | 19 | 6 | 3 | 3 | 1 | 32 |
| Phone screen | 7 | 11 | 7 | 3 | 1 | 29 |
| Onsite | 13 | 20 | 20 | 0 | 0 | 53 |
A problem reported in two rounds counts in both rows. No onsite problem in this set is an ML pipeline. The three that appeared in live rounds were all phone screens, at Stripe, Disney, and Reddit.
Questions we found
Every row is an AI coding task seen in a phone screen or onsite in 2025 or 2026. Frequency counts the separate dates it was seen in those two years. Rows are newest first.
Stripe
Stripe's live rounds mix integration tasks, where you call APIs and stitch the data together, with debugging tasks inside a small library. Most were reported in both phone screens and onsites, and Bikemap and template lookup keep coming back.
| Question | Round | Frequency | All last seen dates |
|---|---|---|---|
| Complete a Bikemap Route and JSON Integration | Phone screen and Onsite | 13 sightings | 2026-10-21, 2026-10-13, 2026-09-29, 2026-09-14, 2026-09-05, 2026-08-18, 2026-07-28, 2026-01-20, 2026-01-15, 2025-12-17, 2025-12-16, 2025-12-11, 2025-12-05 |
| Repair Redirected Request Body Replay | Phone screen and Onsite | 4 sightings | 2026-09-29, 2026-09-23, 2025-11-12, 2025-11-08 |
| Complete a Reconciliation Integration | Phone screen and Onsite | 3 sightings | 2026-09-23, 2026-01-17, 2026-01-14 |
| Integrate and Enrich Contact API Data | Onsite | 1 sighting | 2026-04-13 |
| Repair Template Lookup, Arguments and Includes | Phone screen and Onsite | 10 sightings | 2026-03-04, 2026-02-20, 2026-02-09, 2026-02-07, 2026-01-16, 2026-01-15, 2026-01-14, 2025-12-09, 2025-12-08, 2025-11-21 |
| Request API Endpoints | Onsite | 1 sighting | 2026-03-04 |
| Repair Dynamic-ID Request Replay | Phone screen and Onsite | 1 sighting | 2026-01-15 |
| Complete a Two-Label Tabular Model Pipeline | Phone screen | 1 sighting | 2025-11-11 |
DoorDash
DoorDash's onsite has you run several small services at once. The refund DAG has to keep running as separate local services with real HTTP endpoints, so the interviewer can watch it work. The live tracking round requires a real AI coding tool, and the interviewer looks at your prompts, why you picked a model, how you check the AI's output, and how you review PRs. The profile task comes with follow-ups on status codes, timeouts, retries, backoff with jitter, and circuit breakers.
| Question | Round | Frequency | All last seen dates |
|---|---|---|---|
| Build a Refund DAG with Local HTTP Services | Onsite | 4 sightings | 2026-10-31, 2026-09-16, 2026-08-15, 2026-08-03 |
| Debug Two Routing Strategies | Onsite | 3 sightings | 2026-10-31, 2026-09-16, 2026-08-15 |
| Integrate Three Profile Services | Phone screen and Onsite | 3 sightings | 2026-10-30, 2026-09-16, 2026-08-15 |
| Render a Card Grid from an API | Phone screen | 1 sighting | 2026-10-30 |
| Build Live Dasher Tracking over Server-Sent Events | Onsite | 1 sighting | 2026-10-30 |
| Build a Dasher Payout Service | Phone screen | 1 sighting | 2026-09-02 |
slice
All five slice tasks come from one onsite report on 2026-10-06. Each is a 75-minute service built from scratch around business rules.
| Question | Round | Frequency | All last seen dates |
|---|---|---|---|
| Build a Concurrency-Safe Movie Seat Booking Service | Onsite | 1 sighting | 2026-10-06 |
| Build a Personal Finance Summary Service | Onsite | 1 sighting | 2026-10-06 |
| Build a Role-Based Student Attendance Service | Onsite | 1 sighting | 2026-10-06 |
| Build a Rolling Spending Limits Service | Onsite | 1 sighting | 2026-10-06 |
| Build a Vehicle-Aware Parking Lot Service | Onsite | 1 sighting | 2026-10-06 |
Apple
Apple's live tasks are short, 30 to 45 minutes. The phone screen was a PyTorch UNet to debug, and the onsite tasks were a Flask model API and two extensions of existing code.
| Question | Round | Frequency | All last seen dates |
|---|---|---|---|
| Debug a UNet Segmentation Model | Phone screen | 1 sighting | 2026-02-25 |
| Complete a DAG Task Pipeline | Onsite | 1 sighting | 2025-10-07 |
| Build a Model Inference API | Onsite | 1 sighting | 2025-10-07 |
| Extend a Three-Bot Rock Paper Scissors Game | Onsite | 1 sighting | 2025-08-14 |
The Walt Disney Company
Disney's rounds are junior level and short. One onsite was pair programming on a Spring Boot app with about 40 minutes of coding. In a phone screen, the candidate cloned a Java project from GitHub and set it up in their own IDE while sharing their screen.
| Question | Round | Frequency | All last seen dates |
|---|---|---|---|
| Display, Sort and Search Movie Titles and Years | Onsite | 1 sighting | 2026-01-12 |
| Naive Bayes Spam Classification in a Notebook | Phone screen | 1 sighting | 2025-12-17 |
| Add an API to a Spring Boot Application | Onsite | 1 sighting | 2025-09-09 |
| Return the Course with the Most Students | Phone screen | 1 sighting | 2025-07-21 |
Scale AI
All three Scale AI tasks are onsite services. The LLM categorizer ends with a follow-up where you discuss sharding and parallel LLM calls, then write the parallel-call code.
| Question | Round | Frequency | All last seen dates |
|---|---|---|---|
| Build a CSV-to-JSON LLM Categorization Service | Onsite | 2 sightings | 2026-08-01, 2025-05-09 |
| Implement a Worker Load Balancer | Onsite | 1 sighting | 2025-12-07 |
| Optimize a Resort Driving Route | Onsite | 1 sighting | 2025-05-26 |
Other companies
The other 37 companies have 1 or 2 live-round tasks each, so they share one table, still newest first, with the company in front of each task. One more task came from a report that didn't name the company.
| Question | Round | Frequency | All last seen dates |
|---|---|---|---|
| Rippling: Schedule Future Employee Changes in an HR API | Onsite | 2 sightings | 2026-10-30, 2026-10-29 |
| U.S. Bancorp: Repair an ATM Interface | Onsite | 1 sighting | 2026-10-30 |
| Citadel: Merge Single-Parent Jobs in an Async Job DAG | Phone screen | 1 sighting | 2026-10-29 |
| Reddit: Complete an Imbalanced Recommendation Feature Pipeline | Phone screen | 2 sightings | 2026-10-28, 2026-10-09 |
| Sierra: Debug a Subscription Retention Agent | Onsite | 1 sighting | 2026-10-25 |
| Atlassian: Apartment Attention and Booking Metrics | Onsite | 1 sighting | 2026-10-23 |
| Keka HR: Review a Backend Search Endpoint | Phone screen | 2 sightings | 2026-10-22, 2026-10-21 |
| Keka HR: Build an Alert Ingestion Service | Onsite | 1 sighting | 2026-10-22 |
| Agoda: Refactor a Resilient Vendor Search Aggregator | Onsite | 3 sightings | 2026-10-17, 2026-09-25, 2026-09-21 |
| Flipkart: Implement a Distributed Task Scheduler | Phone screen | 1 sighting | 2026-10-16 |
| Sphere: Integrate a State Tax API | Onsite | 1 sighting | 2026-10-14 |
| Upstart: Build a Poll and Voting API | Onsite | 1 sighting | 2026-10-14 |
| PhonePe: Complete a Journey Stage Event Service | Onsite | 1 sighting | 2026-10-10 |
| Applied Intuition: Driving Log Ingestion and Playback Windows | Onsite | 1 sighting | 2026-10-06 |
| NVIDIA: Build a System Metrics Reporter | Onsite | 1 sighting | 2026-10-06 |
| Zomato / Eternal: Build a Paginated Money Transaction History API | Onsite | 1 sighting | 2026-10-06 |
| Persona: Build an Internal Go-Link Service | Phone screen | 1 sighting | 2026-10-06 |
| Netflix: Review and Repair an ML Feature Pipeline PR | Phone screen | 1 sighting | 2026-10-05 |
| Zip: Repair Monthly Expense Attribution | Onsite | 1 sighting | 2026-10-02 |
| BNSF Railway: Implement a Railway Quote Service Layer | Phone screen | 1 sighting | 2026-09-30 |
| Okta: Implement a Paginated GET API | Phone screen | 1 sighting | 2026-09-30 |
| Groww: Build an Extensible Brokerage Charge Calculator | Phone screen | 1 sighting | 2026-09-25 |
| Odoo: Build FIFO Inventory Costing | Phone screen | 1 sighting | 2026-09-25 |
| Groww: Build an Idempotent In-Memory Wallet Ledger | Onsite | 1 sighting | 2026-09-24 |
| Cresta: Build a Conversational Flight Booking Agent | Onsite | 1 sighting | 2026-09-22 |
| Verkada: Build a Camera Communication API | Phone screen | 1 sighting | 2026-09-22 |
| Pinterest: Implement a Product Read-Through Cache | Onsite | 1 sighting | 2026-09-21 |
| Company not named: Filter and Batch Concurrent Logs | Onsite | 1 sighting | 2026-09-17 |
| Zip: Cache Ticket Alerts for Approval | Onsite | 1 sighting | 2026-09-16 |
| Mercor: Repair a Concurrent Read-Through Cache | Phone screen | 2 sightings | 2026-09-15, 2026-09-13 |
| Shopify: Build a Bank Account Utility | Phone screen | 3 sightings | 2026-09-13, 2025-05-05, 2025-04-28 |
| Amazon: Repair Dashboard Issue Creation | Phone screen | 1 sighting | 2026-09-01 |
| Snap Inc.: Build a Persistent Chat Backend | Onsite | 1 sighting | 2026-08-12 |
| Headway: Fixed-Window Rate Limiter Middleware | Onsite | 1 sighting | 2026-08-06 |
| Headway: Repair Insurance Eligibility Status Projection | Onsite | 1 sighting | 2026-08-06 |
| Baseten: Build a Durable Single-Server Key-Value Store | Phone screen | 1 sighting | 2026-08-02 |
| Baseten: Parallelize API Calls with a Thread Pool | Onsite | 1 sighting | 2026-08-02 |
| Decagon: Build a Resilient AI Gateway | Onsite | 2 sightings | 2026-08-01, 2026-04-13 |
| OpenAI: Repair a Deployed Django Rate Limiter | Onsite | 1 sighting | 2026-07-31 |
| Klaviyo: Repair a Stripe Webhook Handler | Onsite | 1 sighting | 2026-03-27 |
| Klaviyo: Build a Weather App Using an API | Onsite | 1 sighting | 2026-03-17 |
| Tesla: Coordinate Parallel Runs Across Two Exclusive Targets | Phone screen | 1 sighting | 2026-01-14 |
| Motive: Integrate Geocoding and Route Services | Onsite | 1 sighting | 2025-12-23 |
| Motive: Build a Vehicle Catalog Analysis API | Onsite | 3 sightings | 2025-12-19, 2025-12-18, 2025-06-21 |
| Rippling: Build an In-Memory Community Q&A API | Onsite | 2 sightings | 2025-11-20, 2025-01-10 |
| Ramp: Single-Screen Tic Tac Toe | Phone screen | 1 sighting | 2025-05-08 |
Patterns across the reports
- You build more than you fix. 55 of the 76 tasks ask you to build something or extend existing code. All 5 slice tasks work this way, as do Groww's wallet ledger and fee calculator, Snap's chat backend, Persona's go links, and Upstart's polls. Fixes still show up, at Stripe, Headway, Zip, Sierra, and OpenAI, but they're the minority.
- Follow-ups push on resilience and scale. DoorDash's profile task asks about status codes, timeouts, retries, backoff with jitter, and circuit breakers. Scale AI's LLM categorizer turns into a talk about sharding and parallel calls, and then you write the parallel code. Mercor asks about stale reads, maps under concurrent access, and eviction. Snap starts in memory and asks you to plan a move to a database, and Netflix asks how feature serving changes with different freshness needs.
- Pagination keeps coming up. Okta's phone screen adds pagination to a GET endpoint as a follow-up. Klaviyo's webhook round ends with an optional chat about paginating the Stripe API. Zomato's onsite asks you to compare offset, page-number, and cursor pagination before you design the cursor.
- You explain while you work. U.S. Bancorp gives you 15 minutes to read an ATM interface's code and say what it does before you fix anything. Pinterest asks for the overall approach once the cache works, and Applied Intuition asks you to demo the app and explain the tradeoffs. Sierra asks how you tell CLI problems from business-logic problems, then closes with a design follow-up. In one Rippling onsite, the interviewer turned down the candidate's plan to add a separate table and a background worker.
- The setup isn't the same everywhere. In a Disney phone screen, the candidate cloned the project into their own IDE and shared their screen, and a Disney onsite was pair programming. Sierra's round ran in CoderPad. DoorDash's refund DAG has to run as separate local services the interviewer can call while you talk.
- AI tool rules depend on the company. DoorDash requires a real AI coding tool and grades how you use it: your prompts, your model choice, how you check the output, and how you review PRs. Upstart and U.S. Bancorp allow AI tools, and Amazon's phone screen asked candidates to debug a backend with AI assistance. Sierra's CoderPad round didn't allow any.
A few practice tips
- Say your plan out loud before you type: the data model, the endpoints, and what you'll test first.
- Keep a short list ready for follow-ups: timeouts, retries with backoff, idempotency, pagination, and caching.
- Practice cloning and running a repo while sharing your screen, in Java with Spring Boot and in Python with Django or FastAPI.
- If AI tools are allowed, practice talking through what you ask the tool and how you check what it writes.
Keep practicing
- Every problem here is in the AI coding question bank, and each problem name above opens its own practice page.
- For online assessments, read the companion article, 2026 AI Coding OAs at 15 Companies.
- Company interview pages put each company's coding, system design, and AI coding questions in one place, for example Stripe, DoorDash, Apple, OpenAI, Rippling, Snap, Pinterest, and Mercor.
- For full round-by-round accounts from candidates, browse the Experience feed.
AI coding rounds by company
- 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.
- Apple: Apple AI coding interview: CoderPad with a sidebar agent, alongside Apple's other recent interviews.
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.
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.