Over the last two years the Microsoft interview has quietly changed shape, and candidates who prep for the version their seniors described tend to walk in surprised. The coding bar is real, but it sits alongside high-level and low-level design, behavioural questions that now reach into AI judgement, and a final 'as-appropriate' (AA) interviewer who judges whether you clear the bar overall rather than round by round.
For most software roles it runs in three stages: a recruiter screen, an online assessment (the Online Technical Screen, usually two LeetCode-style problems on HackerRank), then a loop of three to five rounds. New-grad and IC2 loops were often three 45-minute rounds; SDE-2 (IC3) and senior loops ran four hour-long rounds. Teams like Azure and CoreAI each flavour the loop differently, and a hiring manager often takes an early round personally.
One quirk trips people up: Microsoft does not always run a dedicated system-design round, even at senior level. One senior candidate was told the first three rounds would be pure data structures and the last a holistic evaluation. Design, when it appears, can be high-level or low-level, and more than one person got a design question they were not told to expect, including at new-grad level.
The other recurring theme is the wait. Candidates describe a fast start, then a week or two of silence while the decision winds through team-matching and approvals, and positive feedback does not always convert to an offer. What comes through across the loop is that it scores judgement, the AA round and the AI-aware behavioural layer, as much as raw algorithms.
Walking into this interview soon? GhostPilot listens to your live call, spots the question the moment it is asked, and puts a structured answer on your screen in real time. Try it on your next mock, or grab a $29 Session Pass, no subscription, for the real thing.
See how it worksThe interview process, round by round
- Recruiter screen Call
A recruiter confirms your background, target level and location, and outlines the loop. Candidates note the recruiter's prep notes are a rough guide, not a script: one candidate found the actual rounds differed from the prep material for two of the rounds.
- Online Technical Screen (OA) ~75 min, HackerRank
Usually two LeetCode-style problems, mostly medium, sometimes with one that is implementation-heavy or harder, and all test cases expected to pass. One candidate traced their assessment questions to a LeetCode contest and AtCoder. An intern applicant separately reported receiving a HackerRank for the screen.
- Technical phone screen (role-dependent) 45 to 60 min
Not every pipeline has one, but many do: behavioural questions plus a single coding problem with a senior engineer. Applied-science and data roles lean on coding plus problem-solving scenarios rather than pure data structures.
- The Loop: coding rounds 45 to 60 min each
The core of the loop. Reported problems include a BFS graph traversal, a Course Schedule style dependency ordering, and dynamic programming from medium to hard. Interviewers want the optimal approach explained with its complexity. Candidates going in ask how much clean, production-quality code counts versus simply getting the problem solved; one new-grad got the optimal solution but fumbled the explanation, and another candidate's interviewer seemed unimpressed even though the approach was correct. Rounds usually open with behavioural questions. One candidate's virtual final ran in Codility CodeLive, a pre-event kickoff then two 45-minute interviews.
- The Loop: system and low-level design 45 to 60 min
Design shows up unpredictably. Some loops have a dedicated high-level design round, others fold low-level design into a coding round, and some senior loops skip a dedicated design round, with one candidate told the last round would be a holistic evaluation instead. Candidates were caught out by design at new-grad and IC2 level, so prepare the basics anyway.
- The final round: AA or manager 45 to 60 min
A senior 'as-appropriate' (AA) interviewer closes some loops and weighs whether you clear the bar overall rather than one skill; other loops instead end with a manager round. Expect deeper behavioural probing and a technical task, and increasingly a question about judgement in AI-powered systems even for non-ML roles.
Interview questions candidates report
Online Technical Screen
- Two LeetCode-style problems, often one straightforward and one harder or implementation-heavy
- All test cases expected to pass, typically within about 75 minutes on HackerRank
- One candidate traced a tough second problem to a LeetCode contest and AtCoder
Coding and data structures
- Graph traversal with BFS
- Course Schedule style topological-sort and dependency ordering
- Dynamic programming in the medium-to-hard range
- Linked-list and heap questions, including one reported in a screening round
System and low-level design
- High-level design of a service, sometimes sprung on new-grad and IC2 candidates
- Low-level design folded into a coding round
- One senior candidate prepping the loop asked how deep to go on APIs, database design, scalability and tradeoffs
- One candidate prepping an AI-heavy loop lined up model-serving tradeoffs, evaluation loops and RAG pipelines to study
Behavioural and managerial
- Walk me through a project on your resume and the tradeoffs you made
- One candidate going in asked how heavily the growth mindset and culture questions weigh
- "Tell me about a time you had to evaluate tradeoffs in an AI-powered system"
Applied Science and ML roles
- Implement K-means from scratch, then vectorise it
- Maximum points on a line, handling floating-point precision
- Explore versus exploit: Thompson sampling against epsilon-greedy
- Calibration, imbalanced-data handling, and data-drift detection
- When offline metrics look better than online: why it happens and how to fix it
AI-assisted coding round
- A LeetCode-style problem (for example a load balancer) solved with an AI tool
- Your own IDE with an assistant such as GitHub Copilot or Claude
- Do not paste the full prompt in; paraphrase the problem to the tool
- They score how you prompt, debug and re-prompt, not just the final code
What candidates said about the process
This candidate applied through the careers portal with no referral and got an assessment weeks later: two problems no harder than medium, one implementation-heavy, and solved both with barely ten minutes to spare. The hiring manager ran round one personally, mixing managerial questions with a partly-written C# data structure to extend for edge cases live. A senior engineer then spent 25 minutes on their AI background before a medium-to-hard dynamic programming problem.
Read the candidate's account →After clearing the assessment in February, this SDE-2 hit a four-round loop after the recruiter's week of silence. Round one was a BFS graph problem solved with complexity analysis. Round two was a low-level design with an abrasive, interrupting senior engineer, but they still coded it. Round three went smoothly with time to spare. The manager round closed with a Course Schedule style question.
Read the candidate's account →A data scientist with about three years of experience described a 60-minute phone screen (a Min Stack problem with follow-ups plus behavioural scenarios) then four hour-long ML rounds: implementing K-means from scratch and vectorising it, a max-points-on-a-line problem where handling floating-point precision cost them time before they reached the optimised solution, and question sets on explore/exploit, calibration and drift detection. The recruiter's prep did not match two of the rounds.
Read the candidate's account →One SDE-2 wrote up the AI-assisted coding round now appearing in some loops, noting almost no prior accounts existed. They used the HackerRank editor alongside an AI assistant on a load-balancer problem, told not to paste the whole question in. Already knowing the two-heap approach, they paraphrased and prompted for it; the interviewer cared most about how they questioned, debugged and re-prompted, not the raw solution.
Read the candidate's account →This candidate breezed through the first two rounds, nailing a technical question and drawing praise on behavioural answers, both LeetCode problems they had seen before. The final round was with a manager who seemed unimpressed and cut them off mid-answer. The technical task was almost trivial, a warm-up, but under pressure they overcomplicated it and stalled. Proof the manager round, and nerves, can sink a strong loop.
Read the candidate's account →How to prepare
- Prep behavioural stories as seriously as algorithms. Every round opens with them and the AA or manager round leans hard on them, and strong technical candidates still got rejected on fit. Ready six to eight STAR stories, including one about a tradeoff in an AI or data-heavy system.
- Talk through the optimal approach and state its complexity before you code, and keep narrating as you go. Explanation carries weight: one candidate got the optimal solution but fumbled the explanation, and another's interviewer was unmoved despite a correct approach. Communicating clearly beats a silently-correct solution.
- Do not skip design because the recruiter downplays it. High-level and low-level design questions surfaced at new-grad and IC2 level, sometimes unannounced. Learn the basics of both even for a junior loop so a surprise design round does not blindside you.
- Expect the format to differ from the recruiter's brief. One candidate found the recruiter's prep material did not match the actual rounds for two of their rounds, so prepare across coding, design and behavioural rather than the single track the brief points at.
- If your loop has an AI-assisted round, practise the workflow, not just the answer. Know the problem cold, paraphrase it to the tool rather than pasting it, and rehearse prompting and re-prompting to fix bugs, because that process is what gets scored.
- Brace for a slow, quiet decision. A fast assessment and loop can be followed by weeks of silence through team-matching and an approval chain, and positive verbal feedback is not a guaranteed offer. Keep other processes warm while you wait.
Frequently asked questions
How many rounds is the Microsoft interview?
After a recruiter screen and the online assessment, most loops are three to five rounds. New-grad and IC2 loops were often three 45-minute rounds; SDE-2 and senior loops ran four hour-long rounds. Each pairs a technical task with behavioural questions, and a senior 'as-appropriate' interviewer helps decide overall.
What is the Microsoft Online Technical Screen?
Microsoft's online assessment, usually two LeetCode-style problems on HackerRank inside roughly 75 minutes. Most are medium, sometimes with one implementation-heavy or harder problem, and all test cases are expected to pass. One intern applicant separately reported getting a HackerRank for the screen.
Does Microsoft ask system design questions?
Sometimes, and not always where you expect. Some loops have a dedicated high-level design round, others fold low-level design into a coding round, and some senior loops skip a dedicated design round, with one candidate told the last round would be a holistic evaluation instead. Candidates were caught out at new-grad and IC2 level, so prepare the basics regardless of level.
What is the AA interviewer at Microsoft?
AA stands for 'as-appropriate', a senior interviewer who closes the loop and weighs whether you clear the bar overall rather than judging one skill. It goes deep on behavioural fit and tradeoff thinking, and carries significant weight in the decision. Some loops instead end with a manager round.
What coding questions does Microsoft ask?
Candidates report a BFS graph problem, Course Schedule style dependency ordering, dynamic programming from medium to hard, and linked-list and heap questions. Explaining the optimal approach with its complexity carries real weight: one candidate reached the optimal solution but stumbled explaining it, while another's interviewer looked unimpressed despite a correct approach.
How long does the Microsoft interview process take?
Often slow after a fast start. Applicants describe an assessment and loop within days, then a week or two of quiet while it winds through team-matching and approvals. Strong feedback does not always convert to an offer, so keep other options open.
Microsoft will ask its own version of these. Paste your actual job description into the free Question Predictor and get the 20 questions that role is most likely to ask, with what each one is really probing.
Predict my questions