Patronus AI
Website:
patronus.ai
Job details:
About Patronus AI
Patronus AI is a frontier lab developing simulation research and infrastructure to accelerate progress toward human-aligned AGI. We are on a mission to simulate all of the world’s intelligence.
We are the team behind some of the earliest and most influential research in AI evaluation like FinanceBench, Lynx, SimpleSafetyTests, CopyrightCatcher, Humanity’s Last Exam, and more. We are formerly AI researchers and engineers from companies like Meta AI, Amazon AGI, and Google. Our customers include foundation model labs and Fortune 500 enterprises like Adobe. We are backed by top-tier investors like Lightspeed Venture Partners, Notable Capital, Stanford University, Noam Brown, Gokul Rajaram, and more.
Responsibilities
As an Engineer at Patronus AI, you will build high-quality simulations used to train, evaluate, and improve AI agents. These environments model real-world applications and workflows, producing the interaction data frontier labs need to improve their models.
You will own complex agent environments end-to-end, working with Applied Researchers to turn specifications into robust software, subject-matter experts to deepen realism, and platform engineers to improve the tooling we build on.
Your work will help frontier labs stress-test and improve the next generation of AI agents, advancing progress toward safe, human-aligned general intelligence.
In this role, you will:
- Build RL environments end-to-end — frontend interfaces, backend services, APIs, data models, and workflows — using technologies like React/TypeScript, Next.js, Python, and SQLite.
- Own delivery of the environments, working closely with Applied Researchers, subject-matter experts, and QA specialists to ensure our simulations are realistic and recognize when something feels wrong in a simulation. Catch seed data that is suspiciously tidy, a workflow with no contradictions, or a UI with none of the accumulated scar tissue of real software.
- Maintain a high bar for the internal tooling and platforms you work with, identifying where it falls short and contribute back to it.
- Think critically about task coverage, environment correctness, edge cases, and adversarial agent behavior, while defining requirements for environments.
- Be a power user of AI coding tools (Claude Code, Codex, and similar), with ideas for extending them and building new tooling on top.
Qualifications
"The number one qualification to succeed in this machine learning course is gumption” - John Lafferty, CS Professor at Yale
Above all, we look for a proactive mindset, willingness to learn, unlimited energy, and relentless optimism. You are a great fit if you have a background in the following:
- B. Tech, M. Tech, or PhD in Computer Science, Machine Learning, Statistics, Mathematics, or a related field with 3+ years of experience as a software engineer, full-stack engineer, product engineer, or in a similar role.
- A track record of owning complex products end-to-end, from ambiguous requirements through production launch, ideally at a startup or in a high-velocity team in a larger company.
- Experience with real enterprise SaaS tools, whether building them or using them deeply in your day-to-day work. CRMs, ticketing systems, billing platforms, or similarly complex applications. You understand how this software behaves after years of users, data, exceptions, and operational constraints.
- Strong engineering judgment around correctness, failure modes, edge cases, and adversarial behavior.
Nice to have:
- Experience with reinforcement learning, agent evaluation, verifiers, reward models, or ML infrastructure — though deep applied research experience isn't required.
- Experience with Playwright, Selenium, or browser automation.
To support close collaboration, this role requires in-office attendance 5 days a week.
Click on Apply to know more.
