Deep, sourced interview guides for the AI companies everyone wants to join. Real processes stage by stage · real questions candidates reported · comp ranges · what gets people rejected.
Frontier labs · Big tech AI orgs · AI-native product companies · AI infra & tooling · Applied AI & forward-deployed
How these are built — each guide is compiled from hundreds of public candidate reports (Glassdoor, Blind, Reddit, interviewing.io, candidate blogs and X threads, 2024–2026), with every process claim and question cited to its source. Anecdote is labeled as anecdote. Processes change — if you interviewed recently, a PR or issue makes the guide better for the next person. ⭐ Star this repo — guides get refreshed and new companies added.
The hardest loops in the industry — research taste, safety, and systems depth.
 OpenAI One of the fastest loops among frontier labs, sometimes two weeks start to finish, testing whether you can architect full-stack AI systems at massive … |  Anthropic The longest loop of the six frontier labs (7-9 rounds), and the dedicated values round is a hard gate: mission mismatch is the most publicly … |  Google DeepMind The most research-flavored loop of the six frontier labs: paper discussions, math and theory rounds, banned AI tools, and a hiring committee that can … |
 Meta AI The only frontier lab that runs two dedicated ML system design rounds in one loop, and the 2026 bar is tighter than the 2025 hiring-spree era after … |  xAI The fastest and most practical loop of the six frontier labs, often under two weeks, screening hard for shipping intensity and in-person Bay Area … |  Mistral AI A fast 3-5 week loop that is brutally selective (a reported ~22% pass rate), rewarding open-weight and open-source engineering over research pedigree. |
ML breadth/depth rounds, team matching, and the AI premium on comp.
 The first big-tech loop to allow AI tools in coding rounds, but the bar shifts to ML breadth and architectural reasoning: if you can't explain the … |  Microsoft A Super Day of 3-4 back-to-back rounds split between Copilot product intuition and Azure infrastructure depth, where "Growth Mindset" answers count … |  Amazon Leadership Principles are a binary gate here: every technical round doubles as a behavioral exam, and failing a single critical LP can sink an … |
 Apple AIML A notoriously slow 5-7 round loop obsessed with on-device ML: fit a 7B model on an iPhone, never violate privacy, and be ready to defend your C++ … |  NVIDIA The most extreme coding bar in big tech: CUDA kernels, GPU memory hierarchy, and hardware intuition decide the offer, and "Python-only" ML engineers … |  Tesla AI First-principles thinking, real-time C++ vision pipelines, and unapologetic mission intensity: every answer must tie back to the physical world of … |
Work trials, paid projects, and product-sense rounds — shipping beats puzzles.
 Perplexity A non-LeetCode Python screen, a 4-5 round loop, and a founder final that probes whether you actually use Perplexity and can prove "frontier … |  Cursor Expect an AI-authenticity grilling on your .cursorrules, a paid 8-hour remote build, and a recruiter who asks up front if you'll accept six-day … |  Scale AI A HackerRank hits your inbox the moment you apply, speed is graded openly, and interviewers tell you to your face the culture is "pretty like 996". |
 Harvey Heavy data-structure coding (R-trees, red-black rebalancing, min-cost flow) with running code expected, a reported ~20% pass rate, and a signature … |  Glean A reported ~3% pass rate, LeetCode medium/hard coding, and a 5-part table-module take-home where finishing 4 of 5 parts is the bar. |  Sierra The AI-native loop: Plan, a 2-hour Build using the coding agents of your choice, then a Review that grades your agency, judgment, and path to … |
 Notion A 24-72 hour block-editor take-home that the pair-programming round then extends live, capped by OT-vs-CRDT system design and a ~23% pass rate. |
Distributed systems, inference economics, and open-source signal.
 Databricks An "extremely high" bar run by a centralized hiring committee, with interviewers candidates compare to the "top 5%" of anyone they've worked with — … |  Hugging Face A famously asynchronous, fully remote process with a literal open-source contribution review round — a merged PR to a Hugging Face repo is … |  Together AI Deep questions on the engineering behind fast, cheap inference for open models — CUDA kernels, KV caches, speculative decoding, and $/1k-token math … |
 Groq Treat Groq as a GPU shop and you fail the bar — the loop centers on the deterministic, compiler-scheduled LPU, C++-heavy low-level fundamentals, and … |  LangChain No LeetCode grind — candidates get dropped into roughly 500,000 lines of real code with no direction or documentation and are judged on what they … |  Weights & Biases A four-stage loop judged on the full MLOps lifecycle and reproducibility — and post-CoreWeave acquisition, candidates who can't articulate the merged … |
 Pinecone A short 3-4 stage loop with a notably low 18.8% positive candidate experience on Glassdoor — ANN fluency, quantization trade-offs, and cost-per-query … |
Decomposition cases, customer scenarios, and practical rounds.
 Palantir The most distinctive loop in tech: a mandatory decomposition round, a mission-fit debate, and a customer scenario sit alongside coding, system … |  Ramp Build, don't puzzle: a CodeSignal Banking System OA, a hotel-reservation phone screen, then 4 practical onsite rounds - offers typically land 9 days … |  Stripe Stripe's signature is troubleshooting plus integration: alongside algorithms, you debug a broken Stripe API flow and integrate a new payment method … |
 Snowflake Snowflake over-indexes on databases and query internals: even general SWE loops ask SQL and schema design, and Cortex/AI roles probe … |  Salesforce MTS loops have pivoted toward Agentforce: 40+ AI questions on Agentforce, the Einstein Trust Layer, RAG, and Data 360 are now standard alongside … |  Duolingo ML here is applied, not research: expect "design a system to improve retention", OO pair programming, and a presentation round - product-ML framing … |
 ServiceNow Scenario-based over live coding: workflow design, Glide scripting, and governed Now Assist/Otto AI agents dominate - "nothing as intense as … |
Interviewed at one of these companies — or one we don't cover yet? Open an issue → with what the loop looked like and any questions you were asked, or open a PR editing the company's file in guides/. Fresh first-hand reports are what keep these accurate. See CONTRIBUTING.md.
1. Company AI context — who's hiring and for what · 2. The loop, stage by stage · 3. Real reported questions (cited) · 4. Technical topics for this company · 5. Design/practical round themes with worked outlines · 6. Behavioral & culture · 7. Compensation · 8. What gets people rejected · 9. Insider tips
Part of the Landed AI-native job-search family:
- 🧭 awesome-ai-native-jobs — the umbrella that maps the whole AI-native job landscape
- ❓ ai-interview-questions — 331 real interview questions with answers
- 🔥 whos-hiring-in-ai — real hiring posts from founders on X
- 💸 recently-funded-ai-startups-hiring — fresh-capital startups staffing up now
- 🚀 ai-engineer-jobs — 300 live AI engineer roles, auto-updated
- 🤝 forward-deployed-engineer-jobs — FDE & customer-facing engineering
- 📈 gtm-engineer-jobs — GTM engineering roles
- 🎓 ai-fellowships-and-residencies — 75 fellowships, residencies & programs
- 🧪 projects-to-land-an-ai-job — portfolio projects that actually get you hired
- 📦 ai-engineer-portfolio-projects — 80+ buildable portfolio projects
- 🗺️ ai-product-engineer-roadmap — the AI product engineer roadmap
- 🎯 become-a-gtm-engineer — the GTM engineer roadmap
