Anthropic Interview Questions — Page 14

Anthropic·Software Engineer·Technical Phone Screen

May 2026

Anthropic coding round focused on a single meaty data structures problem. The question had multiple layers and they clearly wanted to see you think through the design before writing anything.

Given a stream of numeric data points grouped into clusters, design a system that supports adding a value to a cluster and efficiently querying both the mode (most frequent value, ties broken by smallest) and the median for any cluster. Walk through your data structures and the complexity of each operation.

“I jumped straight to the median heap setup and nearly forgot the mode entirely for the first few minutes.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

May 2026

Coding screen for a software engineer role at Anthropic. One problem, simulation-style, and I burned way too much time just parsing what the question was even asking. Got to a working solution but the clock ran out before any follow-ups.

You're given a list of instructions (plus, next, jump) that operate on a global accumulator and a program counter. Exactly one line has its 'jump' and 'next' instructions swapped, causing the program to loop infinitely. Find that line, fix it, and return the accumulated value once the program terminates without looping.

“Spent an embarrassing chunk of time convinced the entire instruction set was broken, not just one line.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

May 2026

Phone screen for a Software Engineer role at Anthropic. The coding portion was a duplicate file finder problem, which sounds manageable until you realize there's no helper code and you're manually reimplementing filesystem methods from scratch. Follow-up questions pushed into distributed systems territory. Still waiting to hear back.

Implement a solution to find duplicate files across a filesystem, without any provided helper code or libraries for filesystem operations.
Where could your duplicate file detection solution break down or produce incorrect results?
How would you scale or improve the efficiency of your solution, for example in a distributed environment?

“The problem itself is pretty classic but not having any scaffolding meant I spent a chunk of time just rebuilding basic filesystem traversal stuff.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

May 2026

Anthropic coding interview, got a web crawler problem that looked deceptively clean on the surface. The BFS structure wasn't the hard part, keeping the constraints straight in your head while coding live was.

Given an API that returns all URLs linked from a given URL, build a web crawler starting from a root URL that visits all reachable pages using BFS. Constrain it to the same domain only and avoid revisiting URLs.

“I jumped straight into BFS and got the basic loop working pretty fast, which felt good.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

May 2026

Technical phone screen for a software engineering role at Anthropic. The whole thing centered on one meaty concurrency question that branched out into Python internals and system design tradeoffs. Felt more like a conversation than a quiz, which I wasn't expecting.

What are CPU-bound and I/O-bound workloads, and how do multithreading, async I/O, and multiprocessing each perform for those workload types in Python? Walk through how the GIL factors into the decision, and which model you'd choose for things like web crawling, parsing, and heavy computation.

“This started as a definitions question and turned into a 20-minute rabbit hole.”

View Post

Anthropic·Software Engineer·Onsite - System Design / Architecture

May 2026

System design round at Anthropic for a software engineer role, centered entirely on designing a prompt playground product from scratch. The question was deceptively broad and pushed hard on product thinking, streaming architecture, and scale, all at once.

Design a prompt playground web product where developers can write prompts, tune model parameters, run generations that stream back token by token, save and version their prompts, and copy the resulting API call.
How would you design the backend request path so that tokens stream from the model back to the browser in real time? What transport mechanism would you use and why?
How do you scale this system to millions of concurrent users, given that each generation is a long-lived streaming connection and GPU inference capacity is the real bottleneck?
How would you implement side-by-side comparison of two prompt variations, streaming both concurrently and showing their outputs and costs aligned?
Where can caching actually help in a generative AI product, and where does it break down?
How do you enforce fair per-user quotas and stop a single user from monopolizing inference capacity by running many long generations concurrently?
A generation streams 800 tokens and the connection drops at token 600. What does the client experience, and how do you let it resume without re-running the whole generation?

“This one is bigger than it looks.”

View Post

Anthropic·Software Engineer·Onsite - Behavioral / Leadership

May 2026

Went through a non-technical culture round for a sales rep role at Anthropic. One question but it stuck with me more than I expected.

What does it mean to be 'mission-first' as a value or working principle?

“I rambled about AI safety and caring about the product, which felt thin the second I said it.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

May 2026

Anthropic data engineer interview that went deep on image processing pipelines. The coding part was manageable but the multithreading discussion caught me off guard more than I expected.

Build a small image processing library that handles a stream of images of varying sizes. It should support grayscale conversion, intensity scaling, and geometric resizing (e.g. nearest-neighbor or bilinear interpolation).
How would you parallelize this pipeline, both across images and within a single image? When does adding threads actually help versus hurt performance?
What are the GIL and thread contention concerns if this pipeline is implemented in Python? How would you work around them?
How do you keep the output ordered when processing images concurrently, given that different images may finish at different times?

“I started with grayscale since that's just a weighted channel average, felt confident there.”

View Post

Anthropic·Software Engineer·Onsite - System Design / Architecture

May 2026

Anthropic system design round focused entirely on building an inference API from scratch. The scope was massive and I kept second-guessing whether to go broad or deep. Walked away feeling like I covered maybe 60% of what they actually wanted.

Design an inference API that serves pre-trained ML models over HTTP or gRPC. Walk through the full system: API contract, model loading and versioning, multi-tenant routing, hardware scheduling, autoscaling, caching, monitoring, and rollout strategy.
How does your design change when you're serving small models versus large language models?
What are the latency versus throughput tradeoffs in an inference service, and how do you expose controls for them in the API or scheduler?
How would you handle model versioning and rollout, including canary deployments, in this system?
How would you monitor for model drift in a production inference service?

“This is a monster of a question.”

View Post

Anthropic·Software Engineer·Technical Phone Screen

May 2026

Data science interview at Anthropic that centered on a marketing analytics case. One question, but it had enough layers to keep me busy for a while.

Define the success metrics for a new user acquisition marketing campaign, then design an experiment to decide whether the campaign should be continued or shut down.

“I started with metrics and probably spent too long on the obvious stuff like sign-up rate and CAC before getting to the more interesting ones like activation rate and 30-day retention.”

View Post