Introduction#
Anthropic's loop has an unusual property: the round candidates underprepare for is the one most likely to end their process. Engineers arrive ready for coding and system design, and then spend an hour discussing safety tradeoffs, disagreement, and integrity with someone who is evaluating the answers carefully.
This guide walks the full stage sequence, separates what Anthropic itself states from what candidates report, and explains how to prepare for the round that does not look technical.
Anthropic's careers site confirms a few concrete things: interviews run over Google Meet, technical roles use live coding tools such as Colab and CodeSignal, and you are allowed to look things up during coding — though you are expected to be comfortable with basic syntax and standard libraries without help. Everything about round counts, gating rules, and loop structure below comes from third-party candidate reports, which conflict with each other. Confirm your loop with your recruiter.
At a Glance#
Reports describe the final loop as four to five rounds of roughly 55 minutes each. Some candidates describe it split across two days; others describe a single long day. One recurring claim — worth knowing but not worth treating as fact — is that a clear failure in coding or system design cancels the remaining rounds, culture round included. That is a blog claim, not an Anthropic statement, but it is a reasonable way to think about your risk: do not bank on a strong culture round rescuing a weak technical one.
Round-by-Round Breakdown#
Online Assessment#
A CodeSignal assessment, commonly reported at around 90 minutes. Its position in the sequence moves — some candidates receive it before the recruiter call, some after.
This is the most conventional part of the process and the part most responsive to ordinary practice. Standard data structures, strings, hash maps, and careful handling of edge cases under a clock. Timed-assessment practice transfers directly here in a way it does not transfer to the later rounds.
Recruiter Call#
Thirty minutes on background, level, and motivation. Ask for the stage list, whether your role includes a take-home, and — specifically — what the policy on AI assistance is for your loop. That last question has a different answer in different places and different quarters, and you want it from your recruiter rather than from a forum.
Technical Screen#
Live coding over Google Meet, in Colab or CodeSignal. Anthropic's own guidance that you can look things up is a meaningful signal about the register of these rounds: they are closer to "write real code in a real environment" than to closed-book recall.
What that does not mean is that fluency is optional. The same guidance asks you to be comfortable with basic syntax and standard libraries. Looking up an argument order is fine; needing to look up how to iterate a dictionary is a cost you will feel.
Final Loop: Coding#
Roughly 55 minutes. Candidate reports emphasize two things more than at most companies:
Library fluency. Working comfortably in your language's standard library — collections, iterators, context managers, the concurrency primitives — rather than reimplementing basics by hand.
Concurrency. Threads, async, locks, race conditions, and reasoning about what happens when two things touch the same state. This appears often enough in reports to be worth dedicated preparation.
Implement a bounded worker pool that processes a queue of jobs with a configurable concurrency limit, retries failures with backoff, and shuts down cleanly on signal — in your main language, in under 40 minutes, with tests. Then explain every synchronization point out loud. If you can do that comfortably, you are prepared for the shape of this round.
Final Loop: System Design#
Roughly 55 minutes. Some candidates report a written component — being asked to lay out reasoning in prose rather than only diagramming. That fits a company with a strong internal writing culture, and it changes preparation: practice stating a design decision and its tradeoff in two clear written sentences, not just in arrows on a canvas.
Topics skew toward what the company builds: API serving under heavy load, request routing and queueing, rate limiting and fair-share across customers, batching, retrieval, evaluation pipelines, and cost control on expensive compute.
Final Loop: Culture Round#
The round that decides borderline cases, and the one candidates consistently name as hardest.
It is not a vibe check and it is not standard behavioral STAR practice. Reported territory includes:
- Safety and ethics reasoning. How you weigh a product decision with a plausible harm attached. Not a quiz on AI safety literature — a test of whether you can reason about tradeoffs honestly rather than performing a position.
- Conflict and disagreement. Specifically, times you disagreed with someone with more authority than you, and what you actually did.
- Integrity under pressure. Times you had to say something unwelcome, ship something imperfect and name it, or stop something that was already moving.
- Sustained challenge. The interviewer pushes back on your answer, sometimes repeatedly. Holding a reasoned position while genuinely updating on good arguments is the signal. Collapsing immediately reads as lack of conviction; refusing to move reads as inability to learn.
How to prepare: write out three real stories where you weighed a tradeoff you were uncomfortable with. Then, for each, articulate the strongest argument for the choice you did not make. If you cannot do that, you will not survive pushback on it.
Final Loop: Hiring Manager#
Team fit, scope, and motivation. Specific interest in the work beats enthusiasm about the company.
Timeline and What to Expect#
Reports on timing vary considerably, and Anthropic does not publish a hiring timeline. Two things the careers site does state are worth knowing: the company does not provide feedback on resumes or interviews, and you can reapply after 12 months — or sooner if your experience has changed materially.
Budget four to eight weeks, and plan for scheduling to be the main source of variance.
A Four-Week Plan#
Week 1 — Assessment conditioning. Timed CodeSignal-style sets. The goal is finishing cleanly under pressure, not solving the hardest possible problems.
Week 2 — Concurrency and standard library. Worker pools, async pipelines, backpressure, graceful shutdown. Write them from scratch, with tests, until the synchronization reasoning is automatic.
Week 3 — Design, written and verbal. Practice API-serving and evaluation-pipeline prompts. For each, produce a short written decision record alongside the diagram.
Week 4 — Culture round, taken seriously. Three tradeoff stories, each steel-manned from the other side, rehearsed with someone instructed to push back hard twice.
FAQ#
How many rounds does Anthropic's interview have?#
Public reports describe a resume screen, a CodeSignal assessment, a recruiter call, a technical screen, and a final loop of four to five rounds of about 55 minutes each. The final loop is sometimes split across two days. Counts vary by role and team.
Can I use AI tools in an Anthropic interview?#
Ask your recruiter, and follow their answer exactly. Reports conflict: some describe loops where no AI assistance is permitted unless explicitly offered, and some describe AI-assisted rounds using Claude Code appearing in certain loops during 2026. Anthropic links to AI guidance for candidates from its careers site, so there is official guidance to request. Do not infer a policy from a blog post, including this one.
Can I look things up during the coding round?#
Anthropic's careers page says you can, while expecting comfort with basic syntax and standard libraries. Treat it as permission to check a signature, not as a substitute for fluency.
Is there a take-home?#
For software engineering roles, usually not — reports mostly describe live and online coding. Research scientist, research engineer, and some ML engineering roles are far more likely to include one.
What is the hardest round?#
Candidates most often name the culture round, and it appears to carry real weight in close decisions. The hardest technical round is typically coding, because of the concurrency and library-fluency emphasis.
Does Anthropic give interview feedback?#
No. The careers site states the company cannot provide feedback on resumes or interviews. You can reapply after 12 months, or sooner if your experience changes materially.
How We Compiled This#
Confirmed details — Google Meet, Colab and CodeSignal, the look-things-up guidance, no feedback, the 12-month reapplication window — come from Anthropic's public careers page. Everything about round counts, loop sequencing, gating rules, and round content is synthesized from third-party candidate reports and prep guides published between 2025 and late 2026. Those sources disagree, sometimes sharply, and we have marked the disagreements. None of this is an Anthropic statement about its current process.