This feature enhances the Triage Agent by allowing managers and dispatchers to coach it in real time during active inbox conversations, in the same way they would coach a human technician. Coaching guidance provided during live interactions can then be surfaced after the fact to inform client intelligence and flag new intents, creating a continuous improvement loop between day-to-day operations and long-term automation quality. Problem The Triage Agent is designed to engage users, ask diagnostic questions, and route or escalate issues appropriately. However, managers and dispatchers often recognize mid-conversation that the agent is asking unnecessary questions, missing critical information, or attempting actions that are not appropriate for the end user, such as walking them through tasks they do not have permission or ability to perform. Today, correcting this behavior requires taking over the conversation, which removes the agent from the thread and disrupts the workflow. This differs from how managers normally coach human technicians, where they can add internal notes with tips, corrections, or areas to focus on while the technician continues working. Additionally, these coaching moments frequently reveal gaps in client intelligence or missing intents, but there is no structured way to capture and reuse that insight. Proposed Solution Introduce a real-time coaching capability within the inbox that allows managers or dispatchers to guide the Triage Agent through the existing notes or an agent-specific coaching field, without taking over the thread. These notes would be internal-only and invisible to the end user, but would immediately influence how the agent adjusts its conversation, including what questions it asks, what actions it avoids, and when it escalates. For example, if the agent begins guiding a user through an application installation that the user cannot perform, a manager could add a coaching note instructing the agent to stop installation assistance, gather specific environment details, and then escalate. The agent would continue the conversation, apply the coaching guidance, and collect all required information before handing off to a technician. After the interaction, the system would flag these coaching points as structured signals that can be reviewed by admins. Repeated or high-impact coaching patterns could automatically suggest updates to client intelligence or be flagged as candidates for new or refined intents. This ensures that real-world coaching directly contributes to improving future triage behavior. Benefits This approach mirrors existing management workflows, allowing managers and dispatchers to coach the Triage Agent just as they would a human technician. It reduces unnecessary takeovers, keeps the agent active longer, and ensures escalations are better prepared. Over time, it transforms ad hoc coaching into durable client intelligence and intent improvements, increasing the accuracy and reliability of automated triage. Use Cases A dispatcher adds a coaching note to redirect the agent away from unsupported troubleshooting steps while it remains active in the conversation. A manager reviews multiple tickets where similar coaching was applied and is prompted to create a new intent to handle that scenario automatically. A support team maintains faster resolution times by letting the agent gather complete information before any human intervention is required.
