TAHI Adapts AI Agents via Iterative Human-AI Interaction
September 4, 2026
Test-time adaptation through human-agent interaction (TAHI) uses cross-session user feedback to bridge the gap between population-scale training and individual expertise. The method integrates interaction signals into agent context and weights using an evolving rubric module to crystallize personal success criteria.
HOW THIS AFFECTS YOU
●
builderYou can implement TAHI to provide users with agents that learn their specific, unstated preferences over time.
●
researcherThis provides a framework for treating human-agent interaction as a signal for test-time weight adaptation.