2026 AI Coding OAs at 15 Companies: 32 Reported Problems
More and more online assessments hand you a real repo instead of a LeetCode prompt. You fix the bugs, add the feature, and make the tests pass on your own, against a timer. This article looks at every AI coding OA FastPrep collected from 2025 and 2026, 32 problems from 15 companies, to see what each company actually asks.
The short version: Amazon has 15 of the 32 problems, and 13 of those are bug fixes inside a full web app. Outside Amazon, the OAs split between similar repairs, tasks that unlock in levels, and ML pipelines.
Coverage: Project-style AI coding tasks from online assessments, seen in 2025 and 2026: 32 problems at 15 companies, seen 79 times in all. Every sighting falls between 2026-04-06 and 2026-11-03. Phone screens and onsites are left out. Levels are junior (17) and mid level (15), with no senior tasks. Only 6 of the 32 problems say whether the candidate was a new grad or an experienced hire. The reports name the platform for 6 problems: HackerRank (4), CoderPad (1), and CodeSignal (1).
What came up
Every one of these OAs is graded by tests. Most ask you to model data inside a repo you didn't write, and 20 of the 32 involve debugging.
| Topic | Problems | Most recent sighting |
|---|---|---|
| Testing | 32 of 32 | 2026-11-03 |
| Data modeling | 22 of 32 | 2026-11-03 |
| Debugging | 20 of 32 | 2026-11-03 |
| API design | 17 of 32 | 2026-11-03 |
| Concurrency | 2 of 32 | 2026-11-03 |
| Transactions | 3 of 32 | 2026-09-12 |
| Security | 1 of 32 | 2026-06-27 |
A problem can count toward several topics, so the rows add up to more than 32.
Bug fixes lead the mix: 19 problems ask you to repair existing code, 6 to complete or extend it, 3 to build from scratch, 3 are ML pipelines, and 1 asks you to write unit tests. Amazon alone accounts for 13 of the 19 bug fixes.
| Company | Bug fix | Complete or extend | Build from scratch | ML pipeline | Write tests | Total |
|---|---|---|---|---|---|---|
| Amazon | 13 | 2 | 0 | 0 | 0 | 15 |
| Goldman Sachs | 3 | 0 | 0 | 0 | 0 | 3 |
| Anthropic | 1 | 1 | 0 | 0 | 0 | 2 |
| McKinsey & Company | 0 | 0 | 0 | 2 | 0 | 2 |
| Walmart | 1 | 1 | 0 | 0 | 0 | 2 |
| 10 other companies | 2 | 2 | 3 | 1 | 1 | 9 |
The Wallet recurring payments repair is one record shared by Amazon and Goldman Sachs, so it counts in both rows and the totals add up to 33.
Questions we found
Every row below is an AI coding OA seen in 2025 or 2026. Frequency counts the separate dates it was seen, so an old favorite can rank below a newer one-off. Rows are newest first.
Amazon
Amazon's OA usually gives you a working app, often a React or Angular frontend over a Django or Spring Boot backend, plus steps to reproduce and the expected behavior. Then you fix the backend. The tasks take 60 to 120 minutes and split between junior (8) and mid level (7). One package tracking OA came as a C++ repo, and two tasks were labeled as AI-assisted debugging. The MovieDB search and recommendations repair is the most-reported OA in this set, with 16 sightings.
| Question | Frequency | All last seen dates |
|---|---|---|
| Repair MovieDB Search and Recommendations | 16 sightings | 2026-11-03, 2026-11-01, 2026-10-31, 2026-10-25, 2026-09-30, 2026-09-11, 2026-09-09, 2026-08-26, 2026-08-19, 2026-08-18, 2026-08-10, 2026-08-05, 2026-07-31, 2026-07-30, 2026-06-28, 2026-05-18 |
| Repair My Cars Ownership and Saved Edits | 4 sightings | 2026-10-21, 2026-10-14, 2026-10-09, 2026-10-01 |
| Repair Workflow Issue and Sub-Issue Creation | 7 sightings | 2026-10-09, 2026-10-01, 2026-09-08, 2026-08-27, 2026-08-21, 2026-08-18, 2026-06-25 |
| Repair Cruzo Car Reviews | 3 sightings | 2026-09-28, 2026-09-24, 2026-09-14 |
| Repair MovieDB Follows and Notifications | 4 sightings | 2026-09-20, 2026-09-19, 2026-09-02, 2026-08-21 |
| MovieDB Custom List Mutations | 1 sighting | 2026-09-16 |
| Repair Django Ticket Creation and Replies | 1 sighting | 2026-09-15 |
| Repair recurring payments in the Wallet backend (shared with Goldman Sachs) | 5 sightings | 2026-09-12, 2026-09-08, 2026-09-01, 2026-08-24, 2026-07-12 |
| Repair Blog Post Creation | 1 sighting | 2026-09-02 |
| Implement Category and Region Product Browsing | 3 sightings | 2026-08-27, 2026-08-26, 2026-08-08 |
| Repair a Package Tracking Status Aggregator | 2 sightings | 2026-08-27, 2026-08-21 |
| Repair Event Listing Filters, Pagination, and Caching | 1 sighting | 2026-08-27 |
| Repair the Wallet Loan Marketplace | 1 sighting | 2026-08-03 |
| Repair Banking App Role-Based Access Control | 1 sighting | 2026-06-27 |
| Repair Workflow Team Editing and Deletion | 1 sighting | 2026-06-25 |
Goldman Sachs
All three Goldman Sachs OAs are repairs in a backend's model layer. Two fix event validation before save and a held-ticket confirmation flow. The third is the Wallet recurring payments repair, which is one shared record with Amazon, so its dates can't be split between the two companies.
| Question | Frequency | All last seen dates |
|---|---|---|
| Repair recurring payments in the Wallet backend (shared with Amazon) | 5 sightings | 2026-09-12, 2026-09-08, 2026-09-01, 2026-08-24, 2026-07-12 |
| Repair Event Pre-Save Validation | 2 sightings | 2026-09-11, 2026-08-25 |
| Repair the Held-Ticket Confirmation Lifecycle | 1 sighting | 2026-08-24 |
Anthropic
Both Anthropic OAs are mid level, come from its Fellows program, and were reported around the same dates. One has you debug an Extremely Randomized Trees implementation that crashes or scores poorly. The other has you write the request scheduler for an LLM inference engine, level by level. The scheduler OA ruled out AI help, web search, and outside libraries.
| Question | Frequency | All last seen dates |
|---|---|---|
| Inference Engine (Fellow Program :D | 4 sightings | 2026-11-03, 2026-08-12, 2026-08-09, 2026-07-30 |
| Debug Extremely Randomized Trees (Fellow Program :) | 3 sightings | 2026-11-03, 2026-08-12, 2026-07-30 |
McKinsey & Company
McKinsey's two OAs are 90-minute data science tasks on HackerRank. You clean the data, train and evaluate a model, and export predictions for a test set. One scores a bot classifier by F1.
| Question | Frequency | All last seen dates |
|---|---|---|
| Predict Customer Support Rating Outcomes (Data Science OA) | 2 sightings | 2026-10-22, 2026-10-21 |
| Classify Bot and Human User Accounts | 1 sighting | 2026-10-21 |
Walmart
Both Walmart OAs are junior, and both were seen on 2026-09-16. One is a 30-minute Spring configuration task about profiles and default values. The other repairs a movie watchlist flow so both views stay in sync and the button rolls back when a request fails.
| Question | Frequency | All last seen dates |
|---|---|---|
| Configure Spring Profiles and Property Defaults | 3 sightings | 2026-09-16, 2026-09-09, 2026-08-27 |
| Repair Movie Watchlist Synchronization | 1 sighting | 2026-09-16 |
Other companies
The other 10 companies have one OA each, so they share one table, still newest first, with the company in front of each task.
| Question | Frequency | All last seen dates |
|---|---|---|
| Keychain: Build an In-Memory Priority Queue Service | 1 sighting | 2026-10-06 |
| IBM: Repair Event Listing Filters, Pagination, and Cache Invalidation | 3 sightings | 2026-09-26, 2026-09-08, 2026-08-19 |
| Intuit: Amend a Git Commit from Bash | 1 sighting | 2026-09-24 |
| Meta: Build an In-Memory Cloud Storage System | 1 sighting | 2026-08-19 |
| Mercury Insurance: Build a Python CRUD API | 1 sighting | 2026-07-30 |
| Coinbase and Airbnb: In-Memory Database | 1 sighting | 2026-07-27 |
| ByteDance: Test Task Management User Assignments | 1 sighting | 2026-06-25 |
| Stripe: Build an Image Prediction Pipeline | 1 sighting | 2026-05-06 |
| Google: Google Docs Mobile Feature Rollout | 1 sighting | 2026-04-06 |
Patterns across the reports
- Fixing a full app from a bug report. You get steps to reproduce, the expected behavior, and a working app, and you fix whatever is broken underneath. That's 13 of Amazon's 15 OAs, and the same shape shows up at Goldman Sachs, IBM, Walmart, and Google. IBM's event listing repair is close to a twin of Amazon's.
- OAs that unlock in levels. Coinbase and Airbnb's in-memory database has four levels, Meta's cloud storage task has four as well, and Anthropic's inference scheduler is also split into levels. Each level builds on the code you wrote for the last one.
- ML OAs are full pipelines. McKinsey's two tasks and Stripe's image prediction OA all have you prepare data, train a model, and export predictions. Anthropic's randomized trees OA is the exception: the model is already written, and you find its bugs.
- Short tasks at the junior end. Intuit's Bash script that amends a git commit takes about 25 minutes, Walmart's Spring profiles task about 30, and ByteDance's OA only asks you to finish unit tests. These check that you know your tools more than they test design.
- AI tool rules vary. Two Amazon tasks were labeled as AI-assisted debugging, while Anthropic's scheduler OA banned AI help outright. Most reports don't say either way, so read the instructions before you start.
A few practice tips
- For repair-style OAs, run the tests before you read much code, then fix one failure at a time.
- For leveled OAs, get each level passing before you start the next one. Keep the code easy to extend, because later levels change it.
- Have Django, Spring Boot, and a React app ready to run locally, since those three cover most of the stacks named in the reports.
- Check whether AI tools are allowed. If they are, practice breaking a task into steps and reviewing what the tool writes. If not, practice without them against a timer.
Keep practicing
- Every problem in this article is in the AI coding question bank, where you can filter by company and level. Each problem name above opens its own practice page.
- For phone screens and onsites, where an interviewer watches you work, read the companion article, 2025–2026 AI Coding Phone Screens and Onsites at 43 Companies.
- Some companies also have an interview page with their algorithm, system design, and AI coding questions in one place: Goldman Sachs, Anthropic, and Stripe.
- 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: 2025–2026 AI Coding Phone Screens and Onsites at 43 Companies: 76 Reported Problems.
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.