About the Role
Join an early-stage AI team to build the agent runtime and evaluation systems that help an AI coworker act reliably in real team workflows. You will shape when agents take initiative, request approval, or stay quiet.
What You'll Do
Design and build the agent runtime for providing context, selecting tools, requesting approvals, and completing work.
Write and refine system prompts to make agent behavior proactive without becoming noisy or unsafe.
Build evaluation and regression tests for real tasks, tool use, and failure cases.
Improve reliability across long-running tasks, interruptions, and retries.
Implement safeguards such as approval gates and multi-tenant data separation.
Work directly with the founding team and early users to take broad problems from prototype to production.
What We're Looking For
Production experience shipping tool-using LLM agents, beyond demos or thin prompt wrappers.
Hands-on system prompt design, improved through testing and iteration.
Experience building evaluation and regression test harnesses for LLM or agent behavior.
Understanding of safe tool execution, approval workflows, and multi-user data risks.
Strong ownership of broad technical problems and the ability to explain tradeoffs.
Experience with LLM agents, tool use, evaluations, and prompt engineering.
Compensation & Benefits
Founding equity is offered; the percentage and vesting terms are to be confirmed.
Location
Remote, from Serbia, Poland, or another Eastern European country. Local employment or contractor structure and working-hour overlap are to be confirmed.