You pick work that teaches you something. You’re the kind of researcher who reads the paper, questions the setup, builds the missing system, and shows the work. Original thinking matters to you, but it is not enough on its own. You turn ideas into code, experiments, prototypes, and products people can actually use.
What you’ll do
You will work on the core problems that appear when agents need to act on behalf of people and organizations: identity, delegation, permissions, memory, provenance, and auditability. The work starts with unclear questions and ends with systems others can run. You might study how an agent should inherit authority, how permissions should persist across tools, how actions should be recorded and reviewed, or how memory and context should be structured so behavior stays reliable over time.