Zoox Interview Questions: The Complete 2026 Guide to the Math Round Everyone Misses

14 min read

Most guides to Zoox interview questions tell you to grind data structures and brush up on C++. That advice isn’t wrong, but it misses the round candidates actually write home about — and the one the search data says people are most anxious about.

Look at what people type into Google after searching Zoox interview questions, and two of the most common follow-ups are zoox math and problem solving interview questions and zoox math interview questions. That’s unusual. Nobody searches for “Google math interview questions.” Something specific is happening in this loop, and the guides ranking on page one mostly don’t explain it.

What Makes Zoox Interview Questions Different

Zoox is an Amazon-owned robotaxi company building a purpose-built vehicle with no steering wheel. That matters for interviews because the engineering problems are genuinely physical — motion planning, sensor fusion, real-time constraints — and the hiring loop reflects it.

Anyone researching Zoox interview questions should start with the volume of public evidence: Glassdoor has 390 user-submitted interview reports for Zoox, with an average reported hiring timeline of 29 days. The reported stage mix is phone interview 31%, one-on-one 20%, skills test 16%, group panel 9%, presentation 9%, background check 5%, personality test 4%, with drug test and IQ test each at 2%.

That “IQ test” category showing up at all is a clue. It’s rare on big-tech Glassdoor pages, and it is almost certainly candidates trying to classify the math round.

The Math and Problem Solving Round Nobody Explains

This is the part of the Zoox interview questions picture worth reading carefully, because it’s where the public record is richest and the prep guides are thinnest.

At least six separate discussion threads exist specifically about a Zoox math round. One commenter addresses the obvious question of whether it’s a new screening fad: “This has been there since the day zoox started as a company.”

What the round actually covers

The most specific topic list from a candidate discussion is: “Geometry & spatial reasoning, science and logic, algebra and arithmetic, probability.” Another account names “Probability, physics, algebra and math was for me.”

On difficulty, the accounts are reassuring and consistent. One candidate reports: “Yeah, it was pretty easy as long as you have some technical common sense. Interviewer was nice and can give you hints. Was about 5 questions.” The same account adds that an “Average 10th grader shouldn’t have any trouble I think” and that the content goes “up to basic trig.”

Another commenter puts it more bluntly: “If you did AP maths you don’t need to prepare for this round.”

The format, described by someone who recognised it

The single most useful characterisation compares it to an older interviewing era: “Questions are similar to 2010-era Google mathy word problems — make some assumptions, do a bit of math with those, ask if you need an equation, talk through your logic.”

That is a precise description of a Fermi-style estimation problem, and it matches the questions showing up on Glassdoor. A Software Engineer report dated 13 January 2026 logs the question “How would you measure the size of the moon.” Another, dated 29 January 2026, logs “Why is it warmer in the summer.”

Neither of those is a coding question or a probability puzzle. They are reasoning-out-loud problems where the interviewer is watching your assumptions.

The math round is a communication test wearing the costume of a quantitative one.

What the interviewers say they want

One reply in these threads reads as coming from the interviewer’s side of the table, and it reframes the whole round: “Make sure to explain your train of thought. We want to know whether you can work with someone on a hard issue, and come up with creative solutions.”

Another candidate’s advice lands in the same place: “Make sure you go in with a positive attitude and don’t give up. It’s behavioral as much as it is anything else.”

A classic problem named in one thread illustrates the type — a bowling ball in a boat floating on a lake, thrown overboard, and what happens to the water level. (It drops.) The point of that question is not the answer. It is whether you reason about displacement out loud without panicking.

One honest caveat on the prep guides

It would be neater to say no prep guide mentions any of this. That isn’t true, and checking matters more than the tidy version. Interview Kickstart’s Zoox guide does explicitly cover this territory, stating “You can expect riddles and brain teasers (typically math-based),” and offering a worked probability problem about drawing same-coloured cards from two differently-sized boxes.

What the guides consistently miss is the estimation half — the moon-measuring, summer-warmth, Fermi-style questions — and the interviewer-stated emphasis on thinking aloud over arriving at a number. The gap is calibration, not total absence.

Zoox Interview Questions by Round

Reported stages vary by team and seniority, but a recognisable shape appears across accounts.

Recruiter screen and online assessment

An intern report dated 24 March 2025 describes an “Online Assessment (C++, Python)” characterised as LeetCode style, followed by a recruiter call. A separate intern account from 7 November 2023 calls the online assessment “easy.”

One intern report from 20 February 2025 in San Mateo logs a question that catches people off guard in a 30-minute screen: “What do you know about Zoox? And how did you find us?”

The technical trivia round

A candidate preparing for a senior phone screen asked about Zoox interview questions and reported hearing “that they have trivia round and coding.”

The reply worth internalising: “The trivia depends on the org” — and, expanded, “Like I said depends on what team. If you are applying for a C++ heavy team you will get C++. For Python you will get Python…Your knowledge depth will definitely be checked.”

An intern account echoes this from the other direction, describing “Python trivial questions but depends on your team.”

Coding rounds

Here the accounts diverge from generic algorithm practice in an interesting way. An intern report describes the online assessment as LeetCode style but the live coding interview as explicitly “Non-LeetCode style.”

The clearest example comes from a Software Engineer report dated 7 January 2026 in Foster City, which describes a one-hour live coding screen on “C++ lane change maneuver planning” — specifically “a lane change maneuver given other vehicles on road.”

Other reported coding prompts are more conventional. A Senior Software Engineer report from 22 November 2024 in Foster City logs an “Array processing algorithm question, with followup to make it more challenging.” A 2 June 2025 senior report describes “Generic python knowledge and programming exercise related to object oriented programming.” A Software Engineer – Perception report dated 8 August 2026 notes it “was focussed more on the OOP side while coding, and made sure your python fundamentals are good.”

System design and domain rounds

A Senior Software Engineer report dated 30 January 2026 in Foster City lists an unusually detailed set of stages: recruiter screen, work presentation, conflict/problem discussion, data structure coding, state machine design, numerical and cognitive analysis, and embedded software architecture. The question logged is “Design state machines for electromechanical systems.”

Note “numerical and cognitive analysis” in that list — the math round again, described in different words by a candidate who accepted an offer.

A Senior Security Engineer report from 21 August 2026 logs “Design a secure network architecture for containerized workloads in Kubernetes,” confirming that design rounds are tailored to discipline rather than generic.

The onsite and its length

The 29 January 2026 Software Engineer report describes a recruiter call, a technical coding screen, an onsite with four rounds (coding, system design, a round the candidate labels “MAPS”, and a manager round), then a leadership call.

Be prepared for a long loop. One detailed account describes a virtual onsite of six interviews, two at manager level, followed by a director conversation — eight rounds in total, ending in rejection. The same account reports the director “said he only learnt about the interview 15 min ago” and “got disconnected from the zoom call around 6 times.” That is one person’s experience and not necessarily typical, but the round count is corroborated by the stage lists above.

How Zoox Interview Questions Vary by Role

Checking Glassdoor’s separate role-title pages surfaces real variation in Zoox interview questions that a single aggregate page would hide.

Embedded software engineer. Three reports, and two of them are notable for what they don’t contain. A 4 March 2023 report states “NDA cannot disclose any of the questions” and a 22 January 2023 report says “Cannot disclose due to NDA.” A 3 May 2024 phone screen of 10–15 minutes asked simply “Do you have any experience with developing driver code?” The NDA pattern is worth knowing: it partly explains why the public record on some Zoox teams is thin.

Software QA engineer. Average reported timeline of 17 days, shorter than the engineering average. A 26 February 2020 Foster City report describes a five-round onsite and logs a genuinely distinctive prompt: “Test case design for fruit peeler machine.” A 19 December 2021 San Francisco report notes the “QA manager asked a lot of questions about previous job experience and main responsibilities.”

Technical program manager. A prep guide describing the TPM loop states that “The next 30 minutes are dedicated to a maths/puzzle, where you collaborate with a ‘partner’ to solve logic problems like optimising a process or solving a complex algorithm” — independent corroboration that the math round extends beyond engineering titles.

Non-engineering roles. A Strategy Associate report dated 28 July 2026 logs “Mostly strategic questions on pricing and market expansion,” confirming the loop shape changes substantially outside technical tracks.

One methodological note if you’re researching this yourself: Glassdoor’s “Sr Software Engineer” and “Senior Software Engineer” pages return the same three reports. Some role pages are aliases rather than distinct pools, so check whether content actually differs before counting a page as a second source.

Zoox Interview Questions to Practice

These are practice exercises written to match the formats described in the candidate reports above. They are not real Zoox interview questions and are not drawn from any candidate account. Use them to rehearse the shape of the round, not to predict its content.

Because the real Zoox interview questions in this round are reasoning prompts rather than recall problems, the value of practising is entirely in the narration — work through each of these aloud.

Estimation and reasoning practice

• Estimate how many miles a robotaxi fleet would need to drive to encounter a given rare road event once. State your assumptions before calculating.
• A sealed container of water sits on a scale. A fish inside begins swimming upward. Does the scale reading change? Reason out loud.
• Estimate the data volume a vehicle with six cameras generates per hour of driving, then per day across a 500-vehicle fleet.
• Roughly how much total braking energy does a city bus dissipate over one day of stop-start service? Work from assumptions you name.

Geometry and spatial reasoning practice

• Two vehicles approach a four-way intersection from perpendicular directions at constant speeds. Derive the condition under which their paths conflict.
• A sensor has a 120-degree horizontal field of view. How many must be mounted, and at what angles, for full coverage with 15 degrees of overlap between neighbours?
• Given a rectangular vehicle footprint and a turning radius, describe the region the vehicle sweeps through a 90-degree turn.

Coding practice in the reported style

• Given the positions and velocities of surrounding vehicles in adjacent lanes, write a function that returns whether a lane change is safe within a fixed time horizon. (This mirrors the reported lane-change prompt’s shape, not its content.)
• Implement a state machine for a vehicle door that must handle open, closing, obstructed, and locked states, including invalid transitions.
• Process an array of timestamped sensor readings and return the longest window in which values stay within a given tolerance band. Then improve your solution’s complexity.

How to Prepare for Zoox Interview Questions

Four practical conclusions follow from the evidence above, and together they cover most of what preparing for Zoox interview questions actually requires.

Don’t over-prepare the math; do practise saying it out loud. Multiple accounts cap the difficulty at AP-level or “basic trig.” The failure mode is going quiet, not missing a formula. Practise narrating assumptions to another person.

Find out your target team’s primary language before the trivia round. The depth check is real and it is language-specific. Ask your recruiter directly — this is a reasonable question and the answer changes what you revise.

Practise domain-flavoured coding, not only abstract LeetCode. Being handed a lane-change planning problem in C++ is a different experience if you have never written code against a physical scenario before.

Plan for a long loop. With a 29-day average and reports of six-to-eight rounds, treat this as a multi-week commitment and keep other processes alive.

For broader technique that applies across autonomous-vehicle employers, our Waymo interview questions guide documents a closely related pattern, and the coding interview questions by company hub covers the rest of the field.

Zoox Interview Questions: FAQ

The Bottom Line on Zoox Interview Questions

The coding rounds at Zoox are real and they skew toward domain problems rather than pure algorithm puzzles. But the round that differentiates this loop — and the one the search data shows people worrying about — is the math and problem-solving stage, which is less a quantitative exam than a structured test of whether you can reason aloud through an unfamiliar physical problem with someone watching.

The encouraging part is that the bar, by several first-hand accounts, is not high. Candidates who struggle reportedly do so by freezing rather than by lacking mathematics. Prepare by talking through estimation problems with another person, find out your team’s language, and expect the process to take a month.

Zoox publishes its own account of what to expect from its hiring process on the Zoox careers journal, which is worth reading alongside candidate reports.