Skip to main content
    Company Guide

    What the Netflix Interview Actually Tests, and Why It Isn't a Whiteboard Grind

    Updated May 2026
    9 min read

    Netflix barely hires new grads. There's no L3-to-L4 conveyor belt the way there is at Google or Amazon. The company runs a flat ladder, and most engineering roles open at what other firms would call senior or staff. So a Netflix interview isn't built to filter a thousand juniors down to a hundred. It's built to answer one question about a candidate who already has the chops: would we fight to keep this person?

    That single framing, which Netflix calls the keeper test, shapes almost everything about the loop. If you walk in expecting two hours of binary trees and a dynamic programming puzzle, you'll prepare for the wrong interview.

    The keeper test, and what it means for an interview

    Netflix's culture memo describes the keeper test as a question managers ask themselves about every report: "if X wanted to leave, would I fight to keep them?" Or, put another way, "knowing everything I know today, would I hire X again?" The same bar applies the first time they meet you. An interviewer isn't checking whether you can pass a course. They're deciding whether, eighteen months from now, they'd still be glad they argued for you in the debrief.

    That changes the kind of answer that wins. A clean solution to a medium LeetCode problem is table stakes, not a differentiator. The signal Netflix is hunting for is judgment: did you make the right call with messy, incomplete information, and can you explain why. The memo pairs the keeper test with a second idea, "context not control," which it describes as giving people the context to make good decisions instead of trying to control what they do. An interviewer who lives that philosophy will poke at how you decide, not whether you arrive at their preferred answer.

    How the loop is weighted

    I'll say this plainly, and some people will disagree with me: the Netflix loop is more behavioral than technical, and that's the opposite of how candidates usually budget their prep time. The coding and design rounds exist, and they're not soft. But they tend to confirm that a senior engineer is in fact senior. The rounds that decide the offer are the ones about how you work, how you disagree, and how you handle a bad call you already made.

    StageWhat it's really probing
    Recruiter screenLevel fit and motivation. They'll check you've actually read the culture memo, not skimmed a summary.
    Hiring manager callDepth of past work and a first read on judgment. This is where a lot of strong coders get filtered out.
    Technical screenPractical coding and a design discussion. Scale-aware, not puzzle-bait. Confirms you're at level.
    Onsite (4 to 6 rounds)Mostly judgment and behavioral, plus one or two design conversations. Every interviewer's read counts.

    The exact count of onsite rounds moves around by team and by level, so treat the number above as a range, not a promise. I don't have a clean dataset on the precise split, and anyone quoting you an exact "47 percent culture, 53 percent code" figure is making it up. What's reliable is the direction: judgment carries more weight here than at most of the rest of big tech.

    The system design rounds are lighter than you'd expect

    This is the part candidates get most wrong. Netflix runs one of the most-studied streaming architectures on the planet, so people assume the design bar is brutal. In practice the design conversation tends to be a discussion, not a gauntlet. They want to see that you reason about blast radius, failure modes, and trade-offs out loud. Reciting a memorized "design Netflix" answer is a tell that you prepared for a script instead of for the actual problem they hand you.

    If you want a model for the kind of design reasoning that lands well, the write-ups at Hello Interview are closer to the right register than the encyclopedic "design X in 45 minutes" video courses. Netflix wants to watch you think under constraints, not hear you replay a YouTube playlist.

    The pay model is its own thing

    Netflix doesn't do RSU grants, sign-on bonuses, or annual performance bonuses the way the rest of FAANG does. They quote one number, mostly cash, and you choose each year how much of it to convert into stock options. The memo's line is that they "pay personal top of market for the role and location," and they re-benchmark every year rather than locking you into a four-year grant that ages badly.

    Netflix software engineer total comp, per Levels.fyi (mostly cash, you elect any stock split):

    • L5 (senior): median around $538,000.
    • L6 (staff): median around $714,000.
    • Full reported range: roughly $218,000 to $1.22M+ across levels.

    For context on how far above the field that sits: the BLS Occupational Outlook Handbook put the median software developer wage at $133,080 in May 2024, with the occupation projected to grow 15% from 2024 to 2034. A Netflix L5 median is roughly four times the national median for the same job title. The all-cash structure has a real downside, though. If the stock doubles, you don't get the windfall a Meta or Google engineer would, because there's no automatic grant riding the upside for you.

    The questions that decide it

    Behavioral questions at Netflix aren't filler between coding rounds. They're the main event. The good ones are open enough that a rehearsed answer falls apart on the second follow-up.

    • Tell me about a decision you made with bad or missing data. What did you bet on, and were you right?
    • Describe a time you disagreed with a leader. What did you do after the decision went the other way?
    • Walk me through a call you got wrong. When did you realize, and what did it cost?
    • You have context but not permission. How do you decide whether to act?

    Notice the shape. None of these reward a tidy STAR template with a happy ending. They reward someone who can sit in the ambiguity, name the trade-off, and own the parts that went sideways. The candidates who do worst are the ones who narrate wins only. Netflix reads that as someone who can't tell the difference between luck and judgment.

    What we hear from candidates using the copilot

    When candidates run Netflix mock loops with the LastRound AI copilot during live rounds, the pattern we hear most often has nothing to do with code. It's that the behavioral follow-ups keep going one layer deeper than they planned for. People show up with a polished story about shipping a feature, and the interviewer skips the feature entirely to ask why they didn't kill the project sooner. Candidates tell us the moment they stopped defending the outcome and started explaining the reasoning is the moment the round turned around. The ones who treat the behavioral rounds as the technical rounds, with the same prep seriousness, are the ones who walk out feeling like it went well.

    Is it the right place for you

    Honestly, the Netflix model isn't for everyone, and I'd rather you figure that out before the onsite than after. The memo is candid that they model themselves on "a professional sports team, not a family," which means high autonomy, direct feedback, and a real possibility of a generous exit if the fit stops working. If you want a structured ladder, predictable promotion cycles, and a stock grant doing the heavy lifting on your comp, several other companies fit you better. If you'd rather be handed context and trusted to run, this is close to the cleanest version of that bargain in the industry. I don't think there's a universally correct answer. There's only the one that matches how you actually like to work.

    Practising the Netflix behavioral loop?

    LastRound AI coaches you through the judgment and context-heavy rounds in real time, including the follow-up questions that decide a keeper-test debrief.

    Sources: Netflix Culture Memo (keeper test, context not control, "professional sports team, not a family"), Levels.fyi Netflix software engineer salaries, and the BLS Software Developers Occupational Outlook. Round-weighting and follow-up observations come from candidates the LastRound AI team has worked with through Netflix mock loops, not from published Netflix data.

    Mahesh

    Written by

    Mahesh

    Founder, LastRound AI

    Founder of LastRound AI. Writes about AI interview tooling, candidate-side interview strategy, and what we learn from running interview-copilot software across thousands of live interviews.

    View Mahesh's LinkedIn profile →

    Further reading

    Share this post

    Related articles