Summary
As we know that Long-Running tools don't support this , leading to RAG and other tools taking a significant amount of time. For tool invocations, we want agents to proactively message the user before, during, and after the tool finishes, without blocking the conversation stream. This improves UX (no “dead air”), enables progress reporting, and supports cancel/retry.

Motivation / Use cases
Heavy tools: web research, RAG over large corpora, batch CRM actions, ETL, model fine-tunes.
Voice & chat UX: avoid silent gaps; narrate progress (“searching…”, “found 3 candidates…”, “summarizing…”).