How Anthropic interviews
These are the Anthropic loops we have researched. Each one lists the rounds, what they test, the follow-up questions to expect, and how to prepare, with the sources we used.
- Software Engineer
- Senior · 5 rounds · 3 h 40 min
- Research Engineer
- Senior · 3 rounds · 2 h
- Product Manager
- Senior · 7 rounds · 4 h 55 min
What Anthropic says about its interviews
Anthropic states that all interviews use Google Meet. Anthropic Careers
Anthropic states that candidates must not use artificial intelligence assistance in live interviews unless Anthropic says otherwise. Guidance on Candidates' AI Usage \ Anthropic
Anthropic Software Engineer interview
Public Anthropic material confirms Google Meet, live coding tools, permitted information lookup, and skills-based evaluation. Candidate reports suggest a 60-minute coding screen and a final loop covering hiring-manager, coding, values, system design, and another coding round. Reports vary by team and date, so exact ordering, durations, tool permissions, and required stages remain uncertain.
The 5 rounds
Technical Coding Screen
Tests practical implementation, state reasoning, and clear manual verification under time pressure.
Expect: Trace transformation from nested call-stack samples. Recursive event generation from changing function samples. Practical software assignment with algorithmic extension.
Follow-ups: “Add repeated, empty, or deeply nested samples.” “Ask for a lower-memory approach and its tradeoffs.” “Change output ordering and test the state model.”
Hiring Manager Interview
Tests senior scope, project ownership, collaboration judgment, and the ability to explain impact clearly.
Expect: Largest project and personal ownership. Conflict over a technical or product decision. Decision made with incomplete information.
Follow-ups: “Ask what you would change now.” “Probe a decision that created disagreement.”
Final Coding Interview
Tests implementation under changing requirements and senior-level tradeoff reasoning.
Expect: Staged cache requirements with eviction. Synchronous workflow changed to asynchronous. Software assignment with embedded algorithmic work.
Follow-ups: “Add persistence or serialization requirements.” “Change synchronous work to asynchronous work.”
Company Values Interview
Tests ethical judgment, self-awareness, constructive challenge, and alignment with responsible work.
Expect: Ethical conflict with competing responsibilities. Hard feedback or a wrong solution. Mission assumptions and constructive skepticism.
Follow-ups: “Ask what changed in your later behavior.” “Probe the cost of your ethical choice.”
Batching System Design
Tests design judgment for high-throughput request orchestration and reliable service operation.
Expect: Large language model sampling API with batching. Queue and worker design under changing load. Interface evolution with latency and failure constraints.
Follow-ups: “Double traffic and vary request sizes.” “Ask how a worker or model endpoint fails.”
What candidates and sources report
- Anthropic states that all interviews use Google Meet. Anthropic
- Anthropic states that technical roles use live coding tools such as Colab and CodeSignal. Anthropic
- Anthropic states that candidates may look up information during technical interviews, but should know basic syntax and standard libraries. Anthropic
- Anthropic states that interviewers ask about candidate experience and motivation. Anthropic
- Anthropic states that candidates have time to ask questions about Anthropic. Anthropic
How to prepare
- Prepare design tradeoffs: Design a batched large language model API in a document. State workload, interfaces, capacity limits, failure handling, observability, and tradeoffs.
Mistakes to avoid
- Starting with optimization before defining event semantics.
- Claiming correctness without tracing shared prefixes and final state.
- Relying on execution instead of explaining the written code.
- Describing a large project without your decisions.
- Blaming others for conflict or weak outcomes.
Questions to ask your interviewers
- Is the five-round final loop current for Senior Software Engineers outside Safeguards?
- Is a separate technical screen required, or can CodeSignal replace it?
- Which coding tools and artificial intelligence tools are allowed in each assessed stage?
Anthropic Research Engineer interview
Official Anthropic material supports Google Meet interviews, live technical tools such as Colab and CodeSignal, permitted factual lookup, and no live artificial intelligence assistance unless allowed. One low-confidence report places a time-pressured CodeSignal assessment before onsite interviews; another mentions a technical screen and culture interview. Exact Senior Research Engineer level, sequence, count, durations, and onsite content remain unknown.
The 3 rounds
CodeSignal Timed Assessment
Test algorithmic coding, prioritization, and correctness under reported time pressure.
Expect: Time-pressured algorithmic implementation. Constraint changes requiring strategy adjustment. Boundary-case and complexity discussion.
Follow-ups: “Change a constraint and ask how the approach changes.” “Probe time and space costs.”
Live Technical Interview
Test live implementation skill, technical reasoning, experience, and motivation.
Expect: Algorithmic implementation with evolving constraints. Numerical machine-learning implementation. Experience and motivation discussion.
Follow-ups: “Change an assumption and ask what must change.” “Probe numerical or algorithmic failure modes.”
Culture Interview
Assess motivation, ownership, collaboration, and reflection through prior experience.
Expect: Motivation for Anthropic and the role. Technical project ownership. Disagreement or collaboration challenge.
Follow-ups: “Ask for a different example when ownership is unclear.” “Probe tradeoffs, disagreement, and measurable impact.”
What candidates and sources report
- Anthropic states that it conducts all interviews over Google Meet and can support different time zones. Anthropic
- Anthropic states that technical roles use live coding tools such as Colab and CodeSignal. Anthropic
- Anthropic states that technical candidates may look up information during technical interviews, but should know basic syntax and standard libraries. Anthropic
- Anthropic states that technical interviews include questions about candidate experience and motivation. Anthropic
- Anthropic states that candidates must not use artificial intelligence assistance in live interviews unless Anthropic says otherwise. Anthropic
How to prepare
- Practice timed algorithmic coding: Use fixed time splits. Write code, state complexity, and perform a verbal manual dry-run without execution or artificial intelligence assistance.
- Rehearse numerical ML implementation: Practice forward pass, loss, backpropagation, and optimizer logic in NumPy. Check shapes and updates through a manual dry-run.
- Prepare experience and motivation stories: Prepare concise stories that show ownership, decisions, collaboration, measurable outcomes, motivation, and changed behavior.
Mistakes to avoid
- Do not optimize before you establish a correct baseline.
- Do not let time pressure replace explicit assumptions and edge-case checks.
- Do not rely on framework calls when the task tests underlying mechanics.
- Do not claim correctness without checking dimensions, state changes, and updates.
- Do not give broad claims without a concrete example.
Questions to ask your interviewers
- Does the CodeSignal assessment apply to Senior Research Engineer candidates, and is it live or take-home?
- What current role title and level map to this Senior Research Engineer process?
- How many onsite interviews occur, and what does each interview evaluate?
Anthropic Product Manager interview
Official Anthropic material does not publish a non-technical Product Manager loop, level name, or round details. One low-confidence guide suggests six stages: recruiter, hiring manager, product sense, analytical or execution, mission and safety, and leadership or cross-functional. One low-confidence candidate report adds a case study and behavioral storytelling, but gives no format or timing. Anthropic requires live interviews without AI unless explicitly allowed.
The 7 rounds
Recruiter screen
Assess product scope, ownership, motivation, and clear communication before the main loop.
Expect: Career narrative, product scope, ownership, and motivation. Role transition, impact evidence, and mission alignment.
Follow-ups: “Probe personal contribution, scope, motivation, and outcome evidence.”
Hiring manager interview
Assess senior product judgment, team fit, ownership depth, and learning through past decisions.
Expect: Team context, product judgment, and ownership depth. Decision tradeoffs, failure, and changed behavior. Manager-dependent priorities and role expectations.
Follow-ups: “Probe alternatives, disagreement, consequences, and later behavior change.”
Case study
Assess structured product reasoning from a supplied brief. The reported format, timing, and scoring remain unconfirmed.
Expect: Supplied case brief and written recommendation. Assumptions, options, risks, and decision criteria. Recommendation, measurement plan, and next steps.
Follow-ups: “Change a constraint and test whether the recommendation still holds.”
Product sense
Assess product improvement judgment, user understanding, prioritization, and outcome-based thinking.
Expect: Product improvement and user need. Product goals, priorities, and constraints. User value, risks, and success measures.
Follow-ups: “Probe user segment, alternative solution, constraint, and success measure.”
Analytical or execution
Assess metric diagnosis, execution judgment, prioritization, and learning from ambiguous evidence.
Expect: Metric drop and segmented diagnosis. Prioritization under resource constraints. Flat experiment with positive qualitative feedback.
Follow-ups: “Change the segment, data quality, experiment result, or resource constraint.”
Mission and safety
Assess responsible product judgment and alignment with Anthropic’s mission and safety context.
Expect: Safety-sensitive product tradeoff. User harm, misuse, and deployment constraints. Mission alignment and escalation judgment.
Follow-ups: “Probe residual risk, mitigation cost, launch conditions, and escalation owner.”
Leadership and cross-functional
Assess senior influence, stakeholder alignment, roadmap judgment, and leadership through conflict.
Expect: Stakeholder disagreement and resolution. Conflicting roadmaps and peer influence. First 90 days and organizational priorities.
Follow-ups: “Probe resistant peers, conflicting incentives, escalation, and durable alignment.”
What candidates and sources report
- Anthropic states that candidates must complete live interviews without AI assistance unless Anthropic explicitly allows it. Anthropic
- Anthropic states that it uses live interviews to assess how candidates think through problems in real time. Anthropic
- Anthropic states that candidates must complete take-home assessments without Claude unless Anthropic explicitly allows AI use. Anthropic
- Anthropic permits candidates to use Claude or other AI tools to refine resumes and application responses, but requires real candidate experience. Anthropic
- Anthropic encourages candidates to use Claude for interview preparation, including company research, answer practice, and question preparation. Anthropic
Mistakes to avoid
- Give a generic career summary without ownership evidence.
- Overstate knowledge of Anthropic’s unpublished process or level structure.
- Describe team activity without separating your contribution.
- Present failure as another person’s fault or a harmless success.
- Treat the case as fully specified when key facts are missing.
Questions to ask your interviewers
- Which stages and durations apply to this Senior Product Manager team?
- Is the case study live, written, or take-home, and how is it evaluated?
- Which mission, safety, and senior-level signals matter most for this team?
Sources
Research reviewed . Practice exercises are our own; they are not confidential interview questions.
- Anthropic Careers — Anthropic
- Shape how AI meets the world — Anthropic
- Anthropic Jobs — Anthropic
- Anthropic Software Engineer, Safeguards Interview Experience (2026) — Candidate report
- Anthropic Interview | Software Engineer | January 2026 — Independent guide
- Anthropic Senior Software Engineer Interview Experience — Candidate report
- Anthropic Senior Software Engineer Interview Experience - London, England — Candidate report
- Anthropic Software Engineer Interview Questions — Candidate report
- Anthropic Interview Question: Similar to a software development assignment with a bit of leetcode-style problems baked in. — Candidate report
- Jobs \ Anthropic — Anthropic
- Guidance on Candidates' AI Usage \ Anthropic — Anthropic
- Does anyone know what the "Google Hiring Assessment" is and what kind of questions one should expect? Thank you! | Glassdoor Forum — Article
- Google DeepMind Salaries in London — Article
- Advice on transition from STEM PhD to industry - Levels.fyi Community — Article
- Anthropic Jobs | Levels.fyi — Article
- Anthropic Research Engineer Interview — Candidate report
- Anthropic interview process — Candidate report
- Jobs at Anthropic — Anthropic
- Jobs \ Anthropic — Anthropic
- Amazon Web Services Senior Technical Product Manager Interview Experience & Questions | Glassdoor — Article
- Anthropic PM Interview structure | Product Management Career - Blind — Candidate report
- Anthropic PM interview Prep | Product Management Career - Blind — Article
- Anthropic Senior Product Manager Interview Questions (2026) — Independent guide
- Anthropic Product Manager Interview Questions (2026) — Independent guide
Independent preparation using Anthropic as a practice target. Not affiliated with, endorsed by, or sponsored by Anthropic.