Introduction#
Anthropic's coding bank is small and unusually coherent. Across the questions in ShowOffer's Anthropic collection, three families account for nearly everything: concurrent processing, multi-entity CRUD systems of the kind that show up in online assessments, and data processing with a correctness twist.
What is conspicuously absent is competitive-programming difficulty. There is no hard dynamic programming, no advanced graph theory. If you are preparing for Anthropic by grinding hard algorithm problems, you are preparing for a round they do not run.
These are the question families in our own Anthropic bank, grouped by the skill each tests. We are not reproducing anyone's list or suggesting Anthropic publishes one. Anthropic's careers page does confirm that technical interviews use live coding tools such as Colab and CodeSignal, and that you may look things up during coding while being expected to know basic syntax and standard libraries.
The Shape of an Anthropic Coding Round#
1. Concurrent Processing#
The highest-return family to prepare, and the one that best predicts how the live round will feel.
How to practice these: build each one with an explicit concurrency limit, retry with backoff, and clean shutdown — then explain every synchronization point out loud. The two crawler variants are worth doing back to back; the difference between the sequential and parallel versions is exactly the gap the round is probing.
Implement a bounded worker pool that pulls jobs from a queue, caps concurrency at N, retries failures with exponential backoff, and shuts down cleanly without dropping in-flight work. Under 40 minutes, with tests. If that is comfortable, this family is handled.
2. Multi-Entity CRUD Systems#
Several questions in this bank are explicitly online-assessment style. These reward clean, boring, well-organized state management across a handful of entity types — and they punish cleverness.
How to practice these: these problems arrive in levels, each adding a requirement on top of the last. The winning strategy is structural: choose a data model in level one that will survive level four. Candidates who optimize level one in isolation end up rewriting at level three and run out of clock.
Read all the levels before writing anything, if the format lets you.
3. Data Processing With a Correctness Twist#
Awkward input, and a subtle requirement that makes the obvious approach wrong.
How to practice these: state the edge cases before you code. The stack-samples problem in particular is a transformation where the natural first approach mishandles boundaries, and catching that out loud is the whole signal.
4. Fundamentals, Done Cleanly#
Not hard. Scored on fluency and pace rather than insight.
How to practice these: write them idiomatically, without reaching for documentation. Anthropic's stated expectation is comfort with basic syntax and standard libraries — these two problems are precisely where a lack of fluency becomes visible.
5. Agent-Shaped Problems#
A small but telling category.
How to practice this one: think about structured state, tool invocation, and what happens when a step fails or returns something unexpected. Error handling is the interesting part, not the happy path.
A Two-Week Practice Plan#
Coding is only one part of this loop, and public accounts consistently describe the culture round as the hardest and frequently decisive part. If you have two weeks, give at least three days to tradeoff stories — not all fourteen to code.
FAQ#
Do I need hard dynamic programming for Anthropic?#
Based on the distribution of questions in this bank, no. The difficulty lives in concurrency correctness and clean multi-entity state design, not in algorithmic depth. Keep fundamentals warm and spend your marginal hour on a worker pool instead.
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 that as permission to check an argument order, not as a substitute for fluency — and ask your recruiter about AI-assistance policy separately, since that is a different question with a different answer.
What language should I use?#
The two fundamentals problems in this bank are Python-specific, and Python is the common default for these loops. Use whichever language you are genuinely fastest in, but be sure you know its concurrency primitives cold.
How is the online assessment different from the live round?#
The assessment is a timed, level-based CodeSignal-style exercise — closest to the CRUD systems in section 2. The live rounds lean more toward concurrency and discussion. Prepare both; they reward different habits.
Are these the exact questions Anthropic will ask?#
No. They are practice problems in the families this loop draws from. Be skeptical of any list claiming to be the real question set.