engineering
Posted Apr 13Software Engineer, Agent Evaluation and Quality
at cursor
San Francisco, United StatesOn-site
Requirements
- WHAT YOU’LL WORK ON - Designing and building best-in-class AI evaluation system: curated datasets, offline replay, scorers / judges, regression alerts, and dashboards.
- YOU MAY BE A FIT IF - You’ve built and operated evaluation or measurement systems, such as AI evals, experimentation, ranking/relevance, or search quality.
- You stay up-to-date and informed on emerging research and industry trends. - You have strong software engineering fundamentals and enjoy shipping production systems.
Contact
- After, we'll schedule an onsite in our office, where you'll work on a small project, discuss ideas, and meet the team. #LI-DNI
Additional details
- The first step in our journey is to build the best tool for professional programmers, using a combination of inventive research, design, and engineering.
- Our organization is very flat, and our team is small and talent dense.
- We particularly like people who are truth-seeking, passionate, and creative.
- We enjoy spirited debate, crazy ideas, and shipping code.
- ABOUT THE ROLE As a Software Engineer on the Agent Quality team at Cursor, you’ll build the measurement, evaluation, and feedback-loop infrastructure that makes the Cursor core agent reliably better over time.
- This role sits at the intersection of product, data, and engineering: you’ll instrument what matters, help define how we judge quality, build pipelines and tooling to analyze agent behavior at scale, and partner closely with research, product, and infrastructure teams to turn insights into improvements.
- Your impact will compound across every Cursor product built on the shared harness—and across high-stakes decisions around model choice, quality, and cost.
- - Designing feedback loops from real usage: collecting, cleaning, and interpreting user signals to inform model and harness changes.
- - Developing analysis tooling and workflows for debugging agent behavior: deep dives on failure modes, clustering themes, and surfacing actionable insights.
- - Improving reliability and guardrails by making quality measurable and operational: defining “good/bad/degraded” sessions, alerting, and triage primitives.