The Anthropic & OpenAI SWE interview loop, by the numbers

What the software-engineering loop at the two frontier labs actually looks like — stage by stage — reconstructed from 60+ publicly-reported candidate accounts (2024–2026).

The loop, stage by stage

The two loops rhyme but differ in emphasis. Anthropic front-loads mission and values; OpenAI front-loads team fit. The single most consistent finding across accounts: a values / culture round appears in essentially every Anthropic onsite, and it fails more technically-strong candidates than any coding round.

Five things that surprise strong candidates

1. AI tools are banned in live rounds. Anthropic enforces it hardest and reportedly uses models to detect test-gaming. Prep with AI; never solve with it live.
2. Coding is build-from-scratch, not LeetCode. You implement a small system and extend it under observation — algorithm trivia doesn't save you.
3. System design is AI-infra-flavored and math-first. Do the capacity math early and let it drive the design — then keep it simple. Over-engineering is the single most-reported design failure.
4. The values / culture round is the #1 filter at Anthropic. It rewards genuine, skeptical, specific thinking; scripted "STAR" answers and flattery backfire.
5. The question bank is small and well-known — so interviewers perturb problems (change a constraint, add a requirement) to test whether you can operate, not just recall.

Coding: the known families

Across accounts, the same handful of build-from-scratch problems recur. Be fluent implementing these in Python (a real edge here): an in-memory multi-level key-value store, a web crawler, an LRU cache, a stack-trace / sampling-profiler exercise, a tokenizer, and a distributed mode/median problem. Knowing them is table stakes; surviving the perturbation is the test. Write extensible code from level one and manage the clock.

System design: the one rule

The most-repeated piece of advice, almost verbatim across sources: do the math first; design the simplest system that meets the stated numbers; bake safety and limits into the request flow; lead the discussion yourself. Recurring Anthropic prompt themes are infra-shaped — serving LLMs efficiently (batching, queueing, GPU utilization), a high-throughput token service, large-scale retrieval, agentic systems. OpenAI leans more product-shaped (APIs, schemas, graceful degradation).

The values round, and how to prepare

It's reflective and probing — "a time your values were tested," "a belief you changed," "a genuine critique of the company." Follow-ups probe your reasoning and honesty, not tidy outcomes. The candidates who pass build a small set of true stories only they could tell, form a real point of view on AI safety, and read the primary sources the loop expects (Anthropic's Core Views on AI Safety, the Responsible Scaling Policy, and Dario Amodei's essays) to engage critically — not to memorize.

Compensation, directionally

Total comp for senior SWEs at these labs clusters in the mid-six figures and up, equity-heavy. The eye-popping headline numbers mostly reflect equity marked to a high private valuation rather than cash; base salaries cluster much tighter. Treat every public figure as directional self-report, not a quote.

See also: the FrontierLoop field guide overview and the free sample chapter.