Jane Street has begun testing new software-engineering interview formats that include working with large language models and evaluating code, a limited change intended to reflect how development work has evolved. The firm says most interview rounds will continue to use its established collaborative coding format without LLM assistance.
The trial started during the summer for some campus candidates attending on-site interviews. Jane Street describes the experimental rounds as involving less direct code writing and more attention to software design, the quality and structure of existing code, the intention behind an implementation and the use of an LLM to generate code. Not every on-site candidate encounters one of the new sessions.
The company has deliberately withheld detailed question formats because the trial is still changing. Candidates will receive advance notice about what to expect when they attend an on-site interview. Jane Street says it plans to iterate before publishing a fuller description.
Its traditional sessions ask a candidate to work through a programming problem with one or two interviewers while explaining decisions. The format is designed to reveal how the person structures work, solves problems, communicates and collaborates. Those objectives remain in place, according to the firm, even as it experiments with ways to assess a broader set of engineering judgments.
The catalyst is the growing use of LLMs inside Jane Street's own development process. The company says most of its engineers now use the tools daily, not only to produce code but also to read, debug and review it, as well as for tasks outside programming. That makes an entirely tool-free interview less representative of some parts of the job.
At the same time, cheaper code generation shifts attention toward choices that are harder to automate or judge from output volume alone. Jane Street highlights deciding what should be built, selecting abstractions that will remain useful and verifying that an implementation is correct. The experimental interviews seek evidence of those abilities rather than treating an LLM's generated code as the end of the task.
The limited rollout also preserves a controlled comparison with existing interviews. Keeping most rounds in the traditional format allows the firm to retain a process it says has worked well while observing whether the new sessions reveal relevant skills. The announcement does not say that candidates are expected to use AI throughout the interview loop or that conventional programming ability has become secondary.
Jane Street's change illustrates a broader challenge for technical hiring: employers need to assess individual reasoning when candidates increasingly work with tools that can rapidly produce plausible implementations. A design or review exercise can expose whether a candidate recognizes fragile abstractions, hidden assumptions and correctness problems, while an AI-assisted task can show how the candidate directs and checks a model.
For applicants, the immediate impact is modest. Most interviews remain collaborative coding exercises without LLMs, experimental rounds appear only in some campus on-sites, and the firm promises notice before a candidate encounters a new format. The trial will determine how much of that mix changes later.
