Long-Horizon Intelligent Agents
Structured long-term context and appropriately timed proactive assistance.
Ongoing Research Long-Horizon Context Proactive Assistance
My ongoing research explores intelligent agents that maintain structured long-term context and turn it into timely, useful action.
The central challenge is not only remembering more information, but identifying causal structure, tracking evolving goals, and recognizing when proactive assistance is appropriate.
I am particularly interested in representations that separate stable user context, active tasks, latent opportunities, and intervention risk. Technical details and results will be added when they are ready for public release.
Research questions
- How should an agent represent tasks and goals that evolve over days or months?
- How can it distinguish a useful service opportunity from an unnecessary interruption?
- How should long-term evidence, prior actions, and changing world state influence the timing of assistance?