Wayve builds a driving model, not a driving stack. The conventional autonomy approach wires together separate modules for perception, prediction, planning, and control, and leans on high-definition maps of every road the car will use. Wayve calls its approach AV2.0: a single foundation model trained on petabyte-scale fleet video that goes from raw sensor input to driving decisions, with the argument that a learned system generalizes to new cities, sensors, and vehicles in a way a hand-tuned modular pipeline does not. The company was spun out of Cambridge, is led by Alex Kendall, and licenses its AI Driver to carmakers instead of operating a robotaxi fleet itself.
Prepare for this loop now because the hiring is broad and the money is fresh. Wayve closed a $1.2 billion Series D in February 2026, which rises to as much as $1.5 billion once Uber’s milestone-based investment is included, at an $8.6 billion post-money valuation, with Nvidia, Microsoft, Uber, Mercedes-Benz, Nissan, Stellantis, SoftBank Vision Fund 2, Eclipse, and Balderton on the cap table. That followed a $1.05 billion Series C led by SoftBank in May 2024. The plan attached to the money is concrete: robotaxi trials with Uber in London from 2026 and the AI Driver in consumer vehicles from 2027, which means both research and production engineering are hiring at once. If you are weighing several private-company bets, our library of company interview guides helps calibrate, but keep Wayve in its own bucket. Waymo runs its own fleet on HD maps; Applied Intuition sells simulation tooling to autonomy teams; Wayve sells a learned driver to automakers and runs supervised robotaxi trials with a ride-hail partner. The interview reflects that middle position.
Why the questions look the way they do
The product decides the interview. An end-to-end driving model is trained, not assembled, so the ML side of the house cares whether you can reason about training on messy, imbalanced, safety-critical data at scale, not whether you can recite a transformer diagram. The systems side exists because a foundation model is worthless if you cannot move petabytes of fleet video through a training pipeline, or run inference on a vehicle inside a tight latency and power budget. So the coding rounds reward correct, readable code and clear reasoning under time pressure, and the design discussion tends to be open-ended about data, scale, and tradeoffs rather than a canned distributed-systems puzzle. Both sides also screen for genuine interest in autonomy, and the final round is framed around mission and values fit rather than a rubber stamp.
What the loop actually contains
Wayve does not publish its hiring process, and the widely available write-ups are SEO reconstructions rather than firsthand candidate reports, so read the table below as a well-informed approximation drawn from role postings and those secondhand accounts. There is no confirmable Glassdoor difficulty score for the role. Treat the round order as likely, not fixed, and confirm your own itinerary with the recruiter.
| Wayve interview stage | Format and length | What it screens for |
|---|---|---|
| Recruiter screen | ~30-45 min video | Background, track fit (ML/research vs. systems), why autonomy, why Wayve, location and work-authorization fit for the specific req |
| Technical round 1 | ~60 min live, with a technical lead | Hands-on coding, code quality, and how you reason through decisions out loud; for ML roles, often ML fundamentals and modeling reasoning too |
| Technical round 2 | ~60 min | Broader engineering discussion: architecture, data and training at scale, on-vehicle constraints, open-ended problem framing |
| Final alignment | ~45-60 min, often a director | Mission and values fit, how you work in a fast-moving research-driven org, collaboration and ownership |
The reported shape is short and personal, closer to two to three weeks than the month-long super-days at bigger autonomy players, with earlier access to the hiring manager. That cuts both ways: with fewer rounds, each one carries more weight, so the alignment conversation likely counts for more than it would in a longer loop. The secondhand accounts do not separate junior from senior loops, so treat level-specific detail as unconfirmed, but expect more weight on the open-ended design and on how you set technical direction and mentor others as the level rises, and ask the recruiter what your req runs. Because autonomy interviewing rewards domain reasoning over raw puzzle speed, a broader read on how AI-era interviews are shifting is worth a pass before you go in.
If you are on the ML / research track
This is the track that builds the driving models and the world models Wayve uses to train and test them. Wayve publishes its GAIA line of generative world models, which produce realistic driving video from text, action, and video prompts and, in the latest generation, put the AI Driver back in the loop for closed-loop simulation and safety evaluation. You do not need to have built that, but you should be able to talk about why a self-driving team would want a learned simulator at all: real-world edge cases are rare and dangerous to collect, so generating and replaying them is how you get coverage. The kind of thing to be ready for, phrased the way the reconstructed loop suggests rather than as leaked questions: how you would tell whether a driving model regressed on rare, safety-critical cases rather than only on an aggregate metric; how you would handle a long-tailed imitation-learning dataset where the events that matter are the rarest; and where imitation learning breaks down and why closed-loop evaluation catches failures a static test set misses. On format, the postings do not say whether the hands-on round is algorithmic (LeetCode-style), practical ML coding (implement a loss, write a dataloader, debug a training loop), or a take-home, so ask the recruiter which one your loop uses and prepare for that specifically. Either way, keep your coding patterns and time and space complexity sharp, and be fluent in Python and PyTorch, which the postings lean on. What separates candidates is connecting a modeling choice to driving safety rather than to a benchmark number.
If you are on the systems / infrastructure track
This half of the org makes the models trainable and deployable. On the data side that is petabyte-scale ingestion, storage, and pipelines that turn raw fleet video into training sets; on the vehicle side it is running a large model inside real latency, compute, and power limits. Role postings on this track lean on C++ and Python, so be genuinely fluent in whichever the req names, and be ready to write clean, compiling code in a shared editor while you narrate your reasoning. The design round here is the one to rehearse: expect something like “design the pipeline that gets a day of fleet video from the car into a training-ready dataset,” or “how would you serve a model on-vehicle within a fixed latency budget.” Structure the answer around scoping first (ask what is in and out of scope before you draw anything), then the data volumes and where they bottleneck, then the tradeoffs you are choosing between. General system-design interview guides transfer, but ground every answer in throughput, latency, and cost rather than reaching for a generic web-service template.
Work authorization and location
Wayve is headquartered in London, with a US office in Mountain View and additional international sites, so the practical answers differ by req. For a London role, work authorization usually means UK right to work; AI companies in the UK commonly sponsor the Skilled Worker visa, but Wayve does not publish a blanket sponsorship policy, so ask the recruiter whether your specific req is open to it. For a US role, ask about US work authorization and sponsorship the same way. On location, Wayve does not publish a company-wide remote, hybrid, or on-site policy, and autonomy work is often on-site because it sits close to vehicles and hardware, so confirm the arrangement for your exact req rather than assuming. Settle all of this on the first call so you are not running a loop you were never eligible to finish.
How to read the comp
Offers at a private company this size come as base plus equity, and the equity is the hard part to value. For a London role, comp is in pounds, and because the UK has no pay-transparency law there is no posted range to anchor on, unlike a California req where a range may appear on the listing. Public aggregators like Levels.fyi and Glassdoor carry only thin data for Wayve, so treat any single figure as order-of-magnitude and ask the recruiter for the band at your level early, since leveling moves the number more than negotiation does. On the equity, pin down whether you are getting options or RSUs, the strike or grant price, the vesting schedule (four years with a one-year cliff is the common shape), and whether refreshers exist, then remember this is still private stock: paper stays paper until a liquidity event, and Wayve has not signaled a public offering. Put the whole offer through a total-comp calculator to value the equity against the current valuation, and a good salary-negotiation walkthrough covers the mechanics of pushing base and grant.
How to prepare
Prioritize by track. ML and research candidates should spend most of their hours on modeling reasoning tied to driving, imitation and reinforcement learning, evaluation, and enough of the world-model idea to discuss why a learned simulator matters, with a steady pass through algorithmic practice so the coding round is not a surprise. Systems candidates should drill data-pipeline and on-vehicle inference design plus clean C++ or Python in a shared editor. Everyone gets the alignment round, and it is framed around values and mission fit, so bring two or three tight stories about ambiguity, ownership, and working in a research-driven team; if yours tend to wander, the STAR structure pulls them back. Spread the technical prep over a couple of weeks with a lightweight study plan so it does not crowd out the domain reading, and run your resume through a resume checker first, since a resume that quantifies what you shipped in ML or infrastructure reads very differently from a generic one.
Wayve is a research bet with a delivery deadline attached. It hires people who can reason about a learned driver on real, dangerous, long-tailed data, and who care that the thing actually drives, which is the question every round is quietly asking.
Practice the behavioral round:
