Microsoft Interview Process & Rounds
In candidate reports, Microsoft hires software engineers in two main ways. A hiring event is an online assessment followed by four back-to-back virtual rounds in one morning, usually three coding rounds and one system design, with a result within days. Team hiring is one team running its own phone screen and a loop of three to five rounds. Either way, most rounds open with 10 to 20 minutes of behavioral questions or your resume, followed by one coding problem that you run against tests you write yourself. Experienced loops add system design and a hiring-manager or project deep dive round. Questions, round types and even the required language vary by team, and candidates say so again and again: MAI and Copilot teams ask more practical coding, and some teams now run AI-assisted coding rounds. From March 2026, several candidates had interviews cancelled, roles paused or offers stall over headcount.
Part of the full catalog of Microsoft interview questions, covering every round.
AceOffer is not affiliated with, endorsed by, or sponsored by Microsoft. Questions are reconstructed from interview experiences candidates shared, not from Microsoft materials.
Key facts
- •7 distinct round types
- •39 questions reconstructed from 134 candidate reports
- •Reports span Oct 2025 – Oct 2026
- •Refreshed monthly · last updated October 2026
The Microsoft loop, from candidate reports
Reported Microsoft software engineering loops vary by team and by route. Hiring events start with a HackerRank online assessment of two problems, such as a balanced permutation check or the cheapest trip when fuel prices differ by city, then four rounds in one morning: three coding rounds and one system design. Team loops start with a phone screen that often spends time on a past project before one coding problem. Onsite coding is mostly LeetCode-style, with Number of Islands and its variants, LRU Cache, a rate limiter and Merge Intervals reported most, plus object-oriented and concurrency problems such as a per-key ordered task executor. MAI and Copilot teams ask more practical multi-part problems: DNA fragment assembly, stop-token detection over a stream of text, beam search decoding and an in-memory table built from CSV. System design rounds most often cover ticket booking, a URL shortener, re-engagement for inactive Copilot users, an online multiplayer game and log storage. Most rounds open with behavioral questions, many loops include a hiring-manager round or a project deep dive, and some 2026 loops add an AI-assisted coding round.
Microsoft interview timeline
| Step | Typical timing |
|---|---|
| Application to first contact | A day to two weeks (7 reports); one candidate waited nearly a month for an OA |
| OA to result | Next day to two weeks (6 reports), including next-day hiring event invitations and a rejection the next business day |
| Phone screen to next step | Next day to two weeks (4 reports); two candidates waited about a month and six weeks for the onsite |
| Hiring event to result | Next day to a week (7 reports); one candidate was told results come within 48 hours |
| Team loop to result | A day to about three weeks (7 reports) |
| End to end | About two to six weeks in reports that give dates (5 reports) |
One hiring event candidate went from recruiter contact to a verbal offer in about six weeks, even after re-booking one round: the OA a week after contact, the event in early November, and the result a week after the last round. Team loops vary more. One candidate's onsite was booked a month ahead, and another waited one to two months after passing the OA for a phone screen because the team preferred candidates near Seattle. Many candidates say recruiters never answer emails or follow-ups, and several describe interviewers arriving late, not showing up, or ending a one-hour screen after 15 minutes. Some invitations come from outside recruiting staff, and in a few reports the interviewer was not the person named on the invite.
The Microsoft interview process, step by step
1. Recruiter contact and choosing a route
An online application or recruiter outreach, then either a hiring event (OA, then four rounds in one morning) or a team's own process. Team hiring often has no OA.
Hiring event candidates usually applied online or were contacted by a recruiter and then placed in an event. Event slots are fixed: rounds start at 8 or 8:30 in the morning and cannot be moved, though one candidate could choose the week from two fixed days each week. In reports, SDE II is level 61 or 62, Senior is 63 and Principal is 65. Team hiring is run by one team: a reply says team hiring skips the OA, and several candidates went straight to a screen or first interview with no OA. Recruiters vary in how much they tell you. Some MAI teams send a written description of the screen, such as an hour of coding close to MAI's work and not too LeetCode-focused, while other candidates were given only round names or focus areas, or no details at all.
- →Ask which level the role is. One candidate realised after the OA that they had applied at SDE II when they meant Senior.
- →Check the time zone on every invite. One candidate read Eastern as Pacific, missed a hiring event round and had to re-book it the next week.
- →Ask for what you want. One candidate who did not want to relocate asked for remote roles, and the recruiter arranged remote interviews.
- →Read any written prompt closely. MAI prompts name the kind of problem, for example a multi-part problem building pieces of a machine learning pipeline in Python on HackerRank.
2. Online assessment
HackerRank, two problems. 90 minutes in experienced and SDE II reports; 75 minutes in new grad reports.
Twenty-two reports describe two problems on HackerRank. One candidate says the browser goes full screen and HackerRank records tab switches. Difficulty varies a lot: several candidates call one or both problems hard, including dynamic programming and a Dijkstra-with-state problem, while others had two mediums or renamed LeetCode problems. The pool overlaps with Uber's: one candidate met a balanced permutation problem they had also seen in Uber's OA. Hidden tests catch slow solutions, and two candidates passed the visible tests but timed out on hidden ones. The passing bar is unclear: one candidate failed three test cases and still got an onsite, others were waiting after passing 9 or 10 of 15, and replies in one thread say a full score did not guarantee an interview. In 2026, one OA showed AI hints on the second problem, and another candidate's OA had no AI assistant.
- →Aim for the optimal complexity before you submit. Two candidates timed out on hidden tests, and a third only had time for brute force.
- →Prepare graph shortest paths and dynamic programming. Both appear in the OAs candidates call hard.
- →Do not count on the OA to move you forward on its own. In one thread, a candidate with a full score heard nothing for a month, and another was not selected after the OA.
3. Phone screen
45 to 60 minutes on Teams with HackerRank or another shared editor, usually time on your resume or a project, then coding. Some are hiring-manager screens.
Most screens start with your background: anywhere from 15 minutes to most of the call on one project. The coding is often one problem, sometimes two in the time. Some screens are with the hiring manager, who checks your background and may add a coding or design question. Some Azure and infrastructure teams ask problems close to their own work, such as deduplicating storage records or expanding interface names. MAI and Copilot screens most often ask DNA fragment assembly, stop-token detection in a text stream, beam search decoding or a top-K problem, and some are mostly behavioral with 15 to 20 minutes of coding. One MAI manager cut a screen to 35 minutes and still asked a coding and a design question. One phone screen was all technical questions with no coding. Applied scientist screens mix ML fundamentals with a short coding problem.
- →Leave time to run your code. One candidate who finished without running tests was rejected.
- →Have a project ready to discuss in depth. Several screens spent half the hour or more on one project, and one candidate was asked about image generation work their resume did not show.
- →Match your introduction to the job description. One rejected candidate's lesson was to tailor the introduction and stories to the role, since the time is short.
- →Ask what the interviewer wants. One candidate solved their problem in ten minutes and was still rejected, which they put down to a different approach from the one the interviewer had in mind.
4. Onsite: coding rounds
Usually two or three rounds of 45 to 60 minutes, each opening with behavioral questions or your resume, then one problem you run against your own tests.
Most coding is LeetCode-style with follow-ups. The most reported problems are number of islands and its variants (7 reports), LRU cache (6), a rate limiter (5) and merge intervals (4). Candidates write their own main function or test cases and run them; one thread says the interviewers wanted the code run, not dry-run. Follow-ups push toward production: thread safety, 100,000 requests a second, and in one new grad loop every follow-up had to be written out, not just discussed. Some teams add an object-oriented design round, such as a notification service or a shopping cart with discounts, and several ask concurrency problems such as a per-key ordered task executor. Some interviewers require a language: C# in three reports, with C++ or Java allowed in two. MAI and Copilot coding is more practical, such as an in-memory table built from CSV or a batch dataloader, and in three MAI threads the poster or replies say the questions matched or were close to ones reported for OpenAI.
- →Practise writing your own test harness quickly. Two candidates ran out of time before their tests passed.
- →Prepare LRU cache with a thread-safe follow-up, and graph search on grids written iteratively. One candidate was asked what happens to recursive DFS on large input.
- →Ask which language is expected. Three candidates were told to use C#.
- →For MAI and Copilot loops, practise multi-part practical problems that grow with each part.
5. Onsite: AI-assisted coding (some teams)
In some 2026 loops, a coding round where you use an AI coding agent in your own IDE while sharing your screen.
Reports describe several versions. One candidate's rounds were labelled as AI coding agent rounds, and the recruiter's prompt asked for an IDE such as VS Code connected to an agent such as GitHub Copilot in agent mode; in the end both were ordinary LeetCode problems where the AI was optional. Another loop handed over a small project to open with your own agent, find a bug and add a feature to persist data to disk, explaining the AI's output along the way, and also had a separate coding round without AI. In one SDE II loop, AI was required in a design round, and the interviewer focused on how the candidate wrote and refined prompts and judged the output. One phone screen showed AI-generated code for the design just discussed and asked the candidate to find its bugs. An AI infrastructure loop had a vibe coding round, and the candidate had to bring their own setup. Interviewers also ask candidates how they use AI in their work.
- →Set up your IDE and coding agent before the interview. One candidate had no suitable agent installed and says to set one up yourself, since the team does not provide one.
- →Explain what the AI produced and check it yourself. Interviewers in these rounds judge your prompts and your review, not only the result.
- →Ask your recruiter whether AI is allowed in each round. Practices differ by team.
6. Onsite: system design
One round in most experienced loops, the fourth round at a hiring event; some loops have two. Some new grad and hiring event loops had none.
The most reported prompts are a ticket or event booking system (4 reports), a URL shortener (4), a re-engagement system for inactive Copilot or ChatGPT users (3), an online multiplayer game (3) and log storage (2). Others include a chat system, a payment system, a distributed key-value store, typeahead, a web crawler, proximity search, an API gateway and a notification system. Many teams ask about their own domain: a RAID system for a storage team, a resource change database, monitoring lab machines, an AI receptionist, or a memory system for a chatbot. Interviewers drill into scale: serving many regions, absorbing write spikes without dropping requests, partitioning, and readers following a file that is still being written. Some interviewers start from a template of requirements, API and database, or ask you to draw, and some designs end with writing code for part of the design. MAI loops add AI and data design, such as sharding a large embedding table across GPUs.
- →Prepare ticket booking and a URL shortener first; together they account for 8 reports.
- →Know why you choose each component. One rejected candidate says they did not answer well why they chose Kafka over another message queue.
- →Be ready to write code for your design. Three reports end the design discussion with implementation.
- →Read the job description for the team's domain. Many prompts come straight from the team's work.
7. Onsite: hiring manager, behavioral and project deep dive
Behavioral questions in nearly every round, plus a separate hiring-manager, behavioral or project deep dive round in many loops.
Because most rounds open with behavioral questions, you will tell several stories across a loop. Topics in reports include conflict or disagreement (7 reports), your most challenging or proudest project, a missed deadline or being blocked, a requirement change mid-project, a production incident after a release, learning something new, negative feedback and why Microsoft. The project deep dive is detailed: interviewers ask about latency in each layer of your system, the trade-offs behind your decisions, and which parts you built yourself, and one candidate recommends having pull request links ready to share. One MAI loop named its rounds after focus areas such as drive for results, influence for impact and bias for action, each half behavioral. In one MAI loop the candidate thinks the hiring manager decided they were not senior enough.
- →Prepare a set of stories rather than one, since the questions repeat across rounds.
- →Pick a project you owned end to end and can explain layer by layer.
- →Prepare a story about a disagreement with another engineer or lead. It is the most reported behavioral topic.
How the Microsoft loop differs by role
Hiring events (SDE II and Senior)
One SDE II candidate describes the event format from the official guide: a recruiter call, a 90-minute OA of two problems, then four virtual rounds starting at 8 in the morning. Rounds are about an hour with breaks between them, usually three coding rounds and one system design, though some events had no design round. Results came within days. Failing an event blocks you from hiring events for six months, but not from applying to team postings. After a team-hire rejection, candidates say they could still apply to other teams, and one recruiter said there was no cooldown.
MAI (Microsoft AI) and Copilot
MAI and Copilot loops are the most varied. One ran rounds one at a time, where failing one ended the loop. Others hold all rounds in one day; one recruiter refused to split it over two days. Coding is often practical and multi-part, such as DNA fragment assembly, stop-token detection, beam search, an in-memory table from CSV or a dataloader. Design rounds include a re-engagement system for inactive users and a multiplayer game. Some machine learning and data roles get an AI exercise and a data design round. Several candidates say their questions matched ones reported for OpenAI. From May 2026, some MAI candidates had interviews cancelled, and one waited on a verbal offer that never became a written one.
Applied scientist
Applied scientist loops mix ML fundamentals (bias and variance, precision and recall, regularization, bagging versus boosting), a deep dive on your research or projects, case studies tied to the team's product, and ML coding such as k-means from scratch. Hiring-manager screens can include coding or design. PhD intern offers were made at the org level, with the team assigned after accepting. One principal candidate passed but was offered one level lower and had to redo the screen and loop.
New grad
New grad loops in reports are three 45-minute rounds of behavioral questions and coding, with no system design or only a simple one such as a rate limiter. The new grad OA is two problems in 75 minutes.
Interns
Software engineering intern reports describe two back-to-back rounds: a medium coding problem with behavioral questions, and in one report a round on debugging, error handling and refactoring a pull request instead of coding. One Copilot research intern had a pure LeetCode interview.
Senior and principal
Some principal loops in reports start with a hiring-manager screen and include system design, coding and a round with a manager from another team; one ran over two days and included an AI fluency interview. One Senior Azure Storage loop had four rounds of a short introduction and a behavioral question followed by 40 minutes of coding, and was rejected as below level 63. Another senior candidate prepared system design for every round, got only coding and behavioral questions, and was offered level 61.
After the final round
Results come by recruiter call or email, or first appear in the application portal. Candidates watch the portal for signals: a status of completed before a verbal offer, a new posting appearing in their portal, or a not-selected status days or weeks after the loop. A reply in one thread says that when a recruiter goes quiet, the portal usually rejects automatically after about two weeks. One hiring event candidate who passed talked with a hiring manager the next day and then got the offer. Offers can come in below the level you expected: one candidate was offered 61 and another 64 instead of 65, and the second was told the compensation team needed a competing offer before raising it. One candidate's position was cancelled two hours after their last round.
What does Microsoft ask in each round?
Microsoft interviews span 7distinct round types, shown below. Counts reflect distinct questions per round across the loops we’ve indexed.
- •Running your code against your own tests. Several reports say rounds expected a main function or test cases that run, not a dry run, and one new grad loop expected every follow-up written out.
- •Preparing the concurrency follow-up. LRU Cache and the hash map are followed by multi-threaded access, the rate limiter by 100,000 requests a second, and the per-key executor asks for per-key ordering under a global concurrency limit.
- •Taking multi-part MAI problems one part at a time. DNA assembly moves from tags with a known direction to an unknown direction and then several chains, and the CSV table from parsing to selecting columns to WHERE filters.
- •Being ready to code part of your design. Some design rounds end with implementation, such as URL shortener code on HackerRank or the trie behind a typeahead.
- •Recursive DFS on large grids. On the region-size problem, one interviewer asked what is wrong with DFS; the candidate's answer was stack overflow on large input, fixed with a queue.
- •Brute force on the OA. Slower solutions timed out on hidden tests, including on the fuel-cost graph problem.
- •Buffering the whole stream in stop-token detection. One report says text must be returned as chunks arrive, with the stop token possibly split across chunks.
- •Skimming the CSV rules. In the in-memory table problem, quoted values, commas inside quotes and type casting cost candidates the later parts.
How to prepare for the Microsoft interview
1. Ask the recruiter which route, level and rounds you have
Find out whether it is a hiring event or a team loop, which level, which rounds include design, whether AI is allowed, and which language is expected. Recruiters often do not reply later, so ask at the start.
2. Practise HackerRank problems on a 90-minute clock
Two problems, one of them possibly hard: graph shortest paths, dynamic programming, sliding windows and binary search. Write the optimal version, because hidden tests catch slow ones.
3. Drill the most reported coding problems with your own tests
Number of islands and its variants, LRU cache with thread safety, a sliding-window rate limiter and merge intervals come up most. Solve each in 30 minutes and run it against tests you write, then add the follow-up.
4. Prepare classic and team-specific system design
Ticket booking, a URL shortener, log storage, a chat system and a multiplayer game, with scale, regions and write spikes ready for follow-ups. Then read the job description and prepare one design from the team's own domain.
5. Write five behavioral stories and one deep-dive project
Cover a disagreement, a hard project, a missed deadline, a change in requirements and a production problem, and prepare one project you can explain layer by layer, with your own decisions and trade-offs.
FAQ
How long does the Microsoft interview process take?▾
What is a Microsoft hiring event?▾
Is there an online assessment?▾
How many rounds are in the Microsoft onsite?▾
Do Microsoft interview questions depend on the team?▾
Can I use AI in the Microsoft interview?▾
Is there a cooldown if I fail?▾
Is Microsoft hiring, or is there a hiring freeze?▾
How fresh is this Microsoft interview data?▾
Get the full Microsoft catalog
Every question, every candidate-reported follow-up, and what passers actually do. Monthly refresh.