Anthropic Interview Questions — Page 31

Anthropic·Software Engineer·Onsite - Behavioral / Leadership

Jul 2026

Interviewed for a SWE role at Anthropic and got a behavioral question about team culture. Short and a bit underwhelming as far as interview content goes.

How do you contribute to a positive team culture?

“I rambled a bit about code reviews and being available for teammates.”

View Post

Anthropic·Software Engineer·Online Assessment (OA)

Jul 2026

Anthropic AI Safety Fellow interview had an OA-style system design problem that was way more infrastructure-heavy than I expected for a safety-focused role. The whole thing revolved around consistent hashing for GPU request routing, which felt more like a distributed systems eng round than anything safety-adjacent.

Design a request-routing layer using consistent hashing to distribute LLM inference requests across a pool of GPU servers, with the goal of maximizing KV-cache reuse and minimizing tail latency. Your solution should include a hash ring with virtual nodes, a routing function that uses session or prompt prefix to select a server, and correct behavior when servers are added or removed. Also discuss why prompt-prefix locality matters here and what trade-offs come with this approach.

“I knew consistent hashing from a distributed systems course but had never thought about it in the context of KV-cache locality before.”

View Post

Anthropic·Software Engineer·Onsite - System Design / Architecture

Jul 2026

System design round at Anthropic for a software engineer role. The whole session was basically one big rate limiter question that kept getting harder as it went on, from algorithm selection all the way into distributed systems territory.

Design a rate limiter that allows at most N requests per second per user. The required API is allow(user_id, now_ts) returning a boolean. Walk through your algorithm options and justify your choice.
Now extend your rate limiter to a distributed setting. Where do you store state, how do you handle clock skew across nodes, and how do you decide whether to fail open or closed when the store is unavailable?

“They wanted me to actually compare the algorithms, not just pick one and run.”

View Post

Anthropic·Machine Learning Engineer·Onsite - System Design / Architecture

Jul 2026

System design round at Anthropic for an MLE role. The core problem was building a distributed cluster status tracker from scratch, which sounds manageable until the follow-ups pile on and you realize they want a full production-grade answer.

Design a cluster status tracker with methods to record node status updates (possibly out-of-order or duplicated), retrieve the current status of a node, retrieve the status at a specific point in time, and return a summary of node counts by status at a given time. Use last-write-wins conflict resolution by timestamp and nodeId. Aim for O(log n) per operation and support up to 100k nodes and 10M updates per day.
How would you extend the tracker to support range queries, such as getting per-minute node status counts over the last K minutes, efficiently?
How would you make the update and read operations thread-safe under high concurrency? Discuss sharding and locking strategies.

“This one took me a while to structure.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

Jul 2026

Anthropic software engineer interview, coding round focused on a classic cache design problem. Pretty standard topic but they pushed hard on the implementation details and complexity guarantees, so you can't just wave your hands at it.

Design and implement an in-memory LRU cache with get and put operations, both running in O(1) time. Walk through your data structures, edge cases, and complexity.

“I knew the answer going in, doubly linked list plus a hashmap, but explaining WHY you need both took more words than I expected.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

Jul 2026

Interviewed for a software engineer role at Anthropic and got hit with a profiler trace parsing problem that looked deceptively manageable until you start thinking about all the edge cases. The core algorithm part was fine but the follow-up about malformed input and infinite loops is where things got interesting.

Given a time-ordered profiler event log with START and END events for a single-threaded program, design an algorithm to compute inclusive and exclusive running times for each function, return the slowest function by exclusive time, and handle edge cases like recursion, cycles, missing END events, and functions that never return.

“I went straight to the stack-based approach: push on START, pop on END, accumulate exclusive time by subtracting child durations.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

Jul 2026

Interviewed for a software engineering role at Anthropic and got a matrix problem that seemed straightforward at first but kept growing. The follow-ups were where things got interesting.

Given an n x n integer matrix, write a function that returns the sum of its main diagonal. Then: if you get q point updates changing individual cells, design a data structure to return the current trace after each update in O(1) time. Also extend it to handle the anti-diagonal sum, and discuss complexity and edge cases like non-square inputs or very large n.

“The base function was fine, just loop the diagonal.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

Jul 2026

Interviewed for a software engineer role at Anthropic and got two back-to-back coding problems, one on file deduplication and one on image processing. Nothing behavioral, just straight into the code. Felt like a solid technical screen, though the image processing part had some tricky edge cases I didn't fully anticipate.

Given a list of directory records (each containing a path and files with their contents), find all groups of files that have identical content and return their full paths. Only return groups with at least two files.
Implement horizontal flip and box blur (radius 1, using in-bounds pixels only, floor division) on a grayscale image matrix, applying operations in a given sequence.

“Pretty standard hashmap problem once you see it.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

Jul 2026

Anthropic SE interview that was basically a deep dive into GRPO training loops. No coding, just a long back-and-forth debugging session where you have to reason about RL policy gradient mechanics, log-prob alignment, and importance sampling. Harder than it sounds if you haven't actually implemented this stuff.

Walk through a complete GRPO training step from prompt sampling to policy update, explaining how group-based advantages are computed and what stays fixed versus what changes during the update.
Identify three subtle bugs in a GRPO or PPO-style training loop that could cause unstable or incorrect learning, covering areas like log-prob computation, masking, advantage normalization, and policy mixing.
Why would the importance-sampling ratio deviate from 1 in a supposedly strictly on-policy setup, and how would you confirm each cause in the training pipeline?
How does the clipped surrogate objective constrain the importance-sampling ratio, and what does the clip range actually protect against once the ratio is no longer guaranteed to be 1?
If all completions in a group receive the same reward, what advantage do they get and should that group contribute to the gradient at all? What numerical safeguard prevents a divide-by-zero during normalization?
In a real generate-then-train stack where log-prob mismatch between the inference engine and the trainer is unavoidable, what is the principled correction and why is ignoring the mismatch wrong?
How does your reasoning about advantage computation and masking change when rewards are given per step (process supervision) rather than as a single scalar per completion (outcome supervision)?
Should the KL penalty against a frozen reference model be folded into the reward signal or kept as a separate loss term, and what are the practical trade-offs of each approach?

“This one took me a minute to structure.”

View Post

Anthropic·Software Engineer·Onsite - System Design / Architecture

Jul 2026

System design interview at Anthropic for a software engineer role. The question was a single massive open-ended prompt covering basically every dimension of backend architecture you can think of. Felt like they were stress-testing how far you could go rather than whether you'd hit a specific answer.

Design a scalable backend service that handles high-volume network I/O for millions of users, covering requirements, APIs, data models, capacity estimates, high-level architecture, scaling strategies, security, failure handling, observability, and edge cases like partial uploads or slow clients.

“This was the whole interview, basically.”

View Post