SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Anysphere, maker of Cursor (an AI-native code editor), is hiring a Software Engineer for the Reinforcement Learning Data team. The company is building frontier coding agents trained on real user data to automate programming tasks.
In this role, you'll own the data pipeline that trains Cursor's coding agents. Your responsibilities include designing task sets that teach specific agent capabilities, iterating on them based on model traces and evaluations until performance improves. You'll analyze agent behavior traces to identify failure modes and surprising patterns, then build systems to surface similar cases for training. You'll also convert one-off recipes into reusable components—better reward functions, cleaner environments, and tighter data quality standards—that other teams can leverage.
You'll partner closely with the research team to validate whether datasets are actually teaching the intended capabilities. The work sits at the intersection of data engineering, machine learning infrastructure, and software craftsmanship.
Ideal candidates have strong software engineering fundamentals and write careful, fast code. You should enjoy breaking fuzzy capabilities into concrete, measurable tasks. Experience in infrastructure, data systems, or distributed systems is valued; RL experience is a plus but not required. You thrive when working with messy real-world agent behavior and turning it into structured datasets or tools.
Anysphere is a small, flat, talent-dense organization founded in 2022 and headquartered in San Francisco. The team values truth-seeking, passion, creativity, spirited debate, and shipping code.
About Anysphere
AI / Data / Infrastructure — maker of Cursor, an AI-native code editor for software developers.
