2026 Stripe Intern AI Coding: A Prep Guide from 10 Candidate Experiences
- Company
- Stripe
- Reports
- 10 interview reports
- Level
- Internship
- Stages
- Onsite
Practice making a small, verified change with AI, explaining it, and then extending it when the requirements grow.
Coverage: Software engineering interns; AI-assisted and AI-enabled programming rounds. Ten candidate accounts published or updated between July and October 2026, including Australia, the United States and Indian recruiting contexts.
What came up
The useful preparation target is a conversation around working code. Candidates described guiding the assistant, explaining its output and keeping up as the task developed.
| What the accounts mentioned | Reports |
|---|---|
| Explaining decisions, reviewing generated code or identifying its problems | 6 of 10 |
| An AI task described in parts | 5 of 10 |
| Testing or edge-case work in the AI round | 3 of 10 |
| An explicitly stated 45- or 60-minute AI session | 3 of 10 |
| The AI round explicitly taking place in HackerRank | 2 of 10 |
| A lengthy prompt or a task longer than the earlier screen | 2 of 10 |
One candidate can appear in several rows. These counts describe what the accounts mentioned; an unmentioned detail may still have been part of the interview.
What should you expect on the day?
Prepare to read a specification, work with an AI assistant and explain the code you accept. Several candidates had a separate API integration interview in the same virtual onsite. Keep the instructions for those two sessions straight: the Sydney candidate could use an embedded agent in the AI session but was prohibited from using generative AI in integration. In an October 29 follow-up about integration, the October 2 candidate also clarified that Google search was allowed and AI was not.
The setup deserves a quick check before your interview. One completed-loop candidate described a README and a task resembling the earlier team screen, with no existing codebase. Another described an agent inside HackerRank. An October report said any LLM was allowed, without explaining whether that included external tools. Confirm the environment, supported languages and permitted assistants in your own invitation.
The reported sessions were 45 or 60 minutes. Known task counts included three and five parts. Neither the available coding time nor a required completion count should be assumed from the calendar slot alone.
How to work through the round
1. Turn the prompt into a small contract
Before asking for code, state the current input, required output and rules in your own words. Identify anything you need to clarify with the interviewer. A long prompt is easier to handle when you can explain the current part in two or three sentences.
This is worth rehearsing. The Sydney intern described being overwhelmed by a long specification and finishing only one part. Their lesson was about breaking the work down and checking the generated implementation.
For practice, write a short checklist beside the editor:
- What must this part do?
- Which rules decide the result?
- What example can I work through myself?
- Which behavior must still work after the next change?
You do not need a complete architecture before starting. You need a clear enough first step to evaluate what the assistant produces.
2. Give the assistant a bounded job
Let the assistant handle implementation once you understand the behavior you want. Give it the current requirement, relevant code and a clear scope. Review its plan briefly, then keep moving.
One candidate who completed the summer-intern loop warned that the assistant could make the solution unnecessarily complex. They recommended direct requirements and pushing back on excess design. Another candidate described the round as mostly prompting and explaining the resulting code.
Here is example wording for rehearsal:
Summarize the current part and list any assumptions. Propose a short implementation plan using the existing functions where possible. Then implement that part, explain the changes and add focused tests with expected results.
If the response adds a framework or several layers you cannot justify, ask for a simpler implementation. If it silently chooses an ambiguous rule, settle that rule before building on it. Your explanation should make clear which decisions came from the specification and which required your judgment.
3. Review the code where behavior is decided
Read enough to explain what the important functions do, how the data moves and which branches determine the answer. Pick an input and trace it through the generated code. Check that the implementation follows the rule you just described.
Useful review questions include:
- Did the assistant omit a condition?
- Does the implementation preserve the required order or priority?
- Is it changing data that a later step still needs?
- Does the new part preserve the earlier behavior?
These are practice checks, not a list of disclosed Stripe questions.
The September HackerRank account specifically mentioned guiding the AI and finding problems in its code. Another completed-interview commenter recommended explaining functions at a useful level instead of reading every line aloud. Be ready to go deeper when the interviewer asks why a particular branch is correct.
4. Make tests answer a question
Ask the assistant to help write tests, but decide what their expected results should be. Running generated code against generated expectations can repeat the same misunderstanding twice.
For each increment, try to cover:
- A normal case that demonstrates the current requirement.
- A boundary case that could change the result.
- An earlier case that the new requirement might break.
Say what a test proves before running it. When it fails, compare the expected behavior with the implementation, make a targeted correction and rerun the relevant checks.
Testing and edge-case discussion appeared in three accounts. The useful rehearsal is being able to explain why a test matters while the clock is running.
5. Balance review with forward progress
One October candidate came up with their own approach, discussed ambiguity and edge cases, and reviewed the code carefully while explaining it. That consumed enough time that they completed about 1.5 of three AI parts. They still received a hiring-manager invitation within a day, after a strong integration round.
That is a useful reminder to keep working when you are behind. It does not tell us the AI pass threshold. A different intern finished all three AI parts quickly and was still waiting to hear whether they would reach the manager round.
During practice, make a habit of ending each part with a short checkpoint: what works, which tests passed and what changes next. If time is running out, explain the remaining change and its main risks. Give the interviewer a clear picture of the work you have verified.
A practice session you can run this week
Use a business-rule exercise with requirements you can introduce in stages. Two existing Stripe practice options are Invoice / Payment Reconciliation and Email Subscription Scheduler. These are rehearsal choices for the skills above; the ten accounts do not establish either as their AI-round question.
For reconciliation, practise adding ID matching, amount matching and tolerance handling in separate increments. For the scheduler, start with scheduled emails, then introduce plan changes and renewals. Use the exercise's stated rules and expected outputs when checking the assistant.
Try a 45-minute practice session with this budget:
| Time | Rehearsal task |
|---|---|
| First 5 minutes | Read the current requirement, explain it and work through an example. |
| Next 25 minutes | Implement and test successive parts with the assistant. |
| Next 10 minutes | Review a new requirement and check that earlier cases still pass. |
| Final 5 minutes | Explain the design, a correction you made and the remaining work. |
This is a suggested practice budget. Use the actual coding time the interviewer gives you during the interview.
For a second attempt, have a friend act as the interviewer. Let them reveal the next part and ask why you accepted a generated change. Review where your time went afterward: reading, waiting for the model, debugging, testing or explaining. That will tell you what to practise next.
Ten candidate accounts behind the advice
01: A README-based task, without an existing codebase
Completed-loop replies: October 26, 2026. A summer-intern candidate said the task came with a README and resembled the earlier team screen with AI added. They warned about over-engineering and recommended stating requirements directly. A separate post from the same candidate confirmed that they had completed the manager interview and were awaiting the decision.
02: Careful review, about half the AI task completed
Completed-VO update: October 26, 2026; exact interview date unreported. This candidate described proposing their own approach, discussing ambiguity and edge cases, and reviewing generated code while explaining it. They reached about 1.5 of three AI parts, completed all five integration parts and received a manager invitation within a day. Their original October 1 post had been a preparation question; the later update supplied the completed-round account.
03: The AI round felt harder than the other coding
Report date: October 23, 2026 · Location: Cambridge, Massachusetts. This intern described a coding interview, an onsite with coding and AI sessions, then a manager chat. They found the AI session challenging. The review was marked as a declined offer, without a disclosed task or completion count.
04: Three AI parts completed, manager decision pending
Report date: October 20, 2026. The candidate finished three of three AI parts quickly and felt the interviewer was pleased. Integration was shakier: roughly 2.5 of five parts, with an API request problem consuming time and pseudocode offered for the rest. The post asked whether they would reach the manager interview; no final result was supplied.
05: A 45-minute, five-part AI interview
Report date: October 18, 2026 · Context: a BTech internship recruiting process. The AI session followed a separate 45-minute integration interview and a 30-minute break. The candidate said LLM use was unrestricted in their AI session, but they had to explain the choices made during the interaction. They initially underestimated the round and later considered it the hardest. They passed the onsite and eventually updated the post to say they were rejected.
06: Four of five AI tasks, then rejection after the manager chat
Interview date: October 2, 2026 · Report date: October 28, 2026. This candidate described two 60-minute onsite sessions, completing five of five integration parts and four of five parts in an AI-related engineering exercise. They advanced to the final manager conversation. According to the feedback they reported receiving, that conversation was the main reason for the eventual rejection. They did not explain the assistant setup or publish the AI task. In an October 29 reply within the integration discussion, they said Google search was permitted but AI was not.
07: A longer task, with emphasis on prompting and explanation
Report date: October 1, 2026. A different candidate in the October discussion described the AI task as similar to the previous screen but longer. They said the work centered on prompting and explaining generated code. Their later advice was to explain functions at a useful level and use the assistant to create and run tests.
08: HackerRank, AI guidance and code inspection
Completed-VO update: September 18, 2026 · Interview period: September. The candidate described two onsite rounds. The AI session took place entirely in HackerRank and involved guiding the assistant through a task resembling the phone screen while identifying code issues. The other session used a GitHub repository and external APIs. No concrete AI prompt or final outcome was shared.
09: The long specification became the obstacle
Interview period: late May to mid-July 2026 · Location: Sydney. The AI session lasted 60 minutes and used an agent inside HackerRank. The candidate struggled to break down a lengthy specification and finished one part. They later emphasized checking generated code and writing regression tests. The final result was a rejection; their belief that the AI round caused it was their own interpretation.
10: A reasonable feeling about the AI round, with no result yet
Interview period: July 2026 · Report date: July 12 · Location: Australia. This intern completed all three integration tasks, felt reasonably good about the AI coding session and had finished the manager interview. They were waiting for the recruiter to return with a decision. The account did not disclose the AI task or its number of parts.
Before the interview
Know how to run code and tests in the environment you will use. Confirm which assistant is available, then rehearse with the tools you are allowed to use. Practise saying what the current part requires, why the generated code meets it and what your tests establish.
You should be able to finish a small change with a clear explanation of its behavior. Then, when another requirement arrives, you have a working version you can extend and check again.
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.
- 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 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.