Reliable Agents in Real Conditions
Building AI agents that plan and carry out multi-step tasks reliably, and know when to ask a person before acting.
Real tasks are messy: information is missing, connections are slow and mistakes have costs. We study planning, memory and clear human-in-the-loop boundaries, so an agent can proceed when it is confident, ask when it is not, and stop when a task is outside what it should do.
- Planning under uncertainty
- Memory across tasks
- Human-in-the-loop boundaries