A friend of mine interviewed at Google in February and got the same dynamic programming problem I’d seen on a Reddit post from a 2024 Meta loop. The questions recycle more than people admit. What changes year to year isn’t the problem types. It’s the framing, the follow-up depth, and what the interviewer considers a satisfying answer.
This is a rundown of what candidates have reported in 2026 across Google, Meta, Amazon, Apple, and Netflix, based on posts in Blind, Levels.fyi forums, and the Stack Overflow Developer Survey. I can’t verify every report, but the patterns are consistent enough to be useful.
How the loops are structured right now
Most FAANG loops run 4 to 6 rounds. Google typically does 1 phone screen plus 4 onsite rounds (2 coding, 1 system design, 1 behavioral). Meta runs a similar format but adds a product sense or cross-functional interview for senior roles. Amazon almost always includes 2 behavioral rounds focused on Leadership Principles, sometimes more. Apple is the outlier: their loops are shorter on average (3 to 4 rounds), more focused on domain depth, and harder to generalize because the hiring team structure varies so much by org.
Netflix is still unusual. Their interviews are heavily behavioral and cultural-fit focused. The coding bar is genuinely high, but it’s the “dream team” conversation that eliminates most candidates. Many people walk out of a Netflix onsite thinking the coding went well and then get rejected on culture fit. That happens more than people realize.
Coding questions by company
The honest answer is that Google and Meta both use medium-difficulty LeetCode problems most of the time. Hards show up, but they’re not the default. What differentiates Google interviews is follow-up depth. They’ll ask you to optimize an already-working solution two or three times, and the candidate who gets an offer is usually the one who keeps finding improvements rather than stopping at the first accepted answer.
Amazon’s coding rounds are similarly medium-weighted. They care more about clean code and explaining your decisions out loud than hitting edge cases you didn’t think of. Their behavioral bar is where people actually fail, not the algorithm round.
Apple tends to ask domain-specific questions, particularly for iOS and machine learning roles. If you’re interviewing for a Swift-heavy role, expect questions about memory management and concurrency that you genuinely won’t find on generic prep lists.
Topic areas that keep appearing in 2026 candidate reports:
- Arrays and two-pointer patterns (all companies)
- Graphs and BFS/DFS traversal (Google, Meta, Amazon)
- Dynamic programming on subsequences and grids (Google heavy)
- Tree problems with recursive and iterative solutions (all companies)
- Sliding window on strings (Meta, Amazon)
- Bit manipulation (Apple, occasionally Google)
System design: what’s different in 2026
The classic questions are still there. Design a URL shortener. Design Twitter’s feed. Design a distributed cache. But there’s a new category emerging from 2025 onward: AI-adjacent system design. Candidates at Google and Meta have reported questions like “design a vector search system” or “how would you architect a feature store for an ML pipeline at scale.” You don’t need to be an ML engineer to answer these, but you need to know what a feature store is and why latency matters in retrieval.
The LinkedIn Economic Graph research on in-demand skills consistently shows ML infrastructure knowledge rising faster than almost any other category. That’s showing up in system design interviews now.
For senior and staff-level candidates, the evaluation criteria have also shifted. A correct design matters less than how you handle trade-offs. Interviewers want to hear you say “I’d use a relational DB here instead of Cassandra because read patterns are predictable and the team is small.” Decisiveness under uncertainty, with a clear reason, scores better than exhaustive enumeration of options.
Behavioral rounds and what each company actually listens for
Amazon uses the STAR format almost exclusively. They map every answer to one of their 16 Leadership Principles, and interviewers are trained to probe until they find one. The mistake most people make is giving vague answers with no disagreement or conflict. Amazon specifically wants to hear about times you pushed back on a decision or delivered bad news. “I agreed with everyone” is not a useful Amazon story.
Google’s Googleyness interviews are harder to game because they’re looking for things like intellectual humility and comfort with ambiguity. Answers that are too polished tend to feel rehearsed. A bit of genuine uncertainty actually helps here.
Meta’s behavioral questions lean toward speed, prioritization under pressure, and data-driven decision-making. They’ll often ask what you would deprioritize, not just what you’d build.
Netflix asks almost exclusively about your opinions and judgment. “What’s a mistake you made that you’d make differently today?” and “Tell me about a project where you had to set your own direction with no clear guidance” are representative questions. The culture deck they published is still accurate, and reading it before the interview is worth the 20 minutes.
One thing I’d do differently if preparing now
I think the Blind 75 is overrated for senior interviews. It’s fine for new grads practicing pattern recognition, but senior candidates who spend 90% of their prep time on LeetCode and 10% on system design and behavioral stories are getting that ratio backward. For a principal-level loop, the algorithm round might actually be the easiest part of the day.
Mock interviews with real-time feedback help more than solo grinding past a certain point. Tools like Craqly let you run practice sessions with an AI copilot that surfaces relevant hints during the interview without just handing you the answer, which mirrors the kind of nudge you might get from a supportive but not overly helpful interviewer. It’s a decent way to simulate the pressure.
The question you should probably ask yourself right now: are you stuck on the same category of problem over and over, or are you stuck on the behavioral stories you can’t seem to make concrete? Most people know which one it is.