Guided playback running.
Concept and sources

This console combines three patterns that now recur across agent platforms: parallel supervision of live agent conversations, trace-level observability, and a deliberately designed transfer between agent and human. The trust states that scope which tools the agent may use follow the trust-context model voice-agent platforms adopted in 2026, and the approval gate for risky actions follows the action-guard pattern from the Magentic interface project at Microsoft Research. The warm handoff treats escalation as a first-class moment: context is composed for the operator rather than dropped, and the return to the agent closes the loop with a summary — the calibrated-trust findings from a 2025 study on appropriate reliance argue for exactly this kind of visible evidence over bare confidence claims.

The blueprint view adds the design side of the same loop: an interview builds a shared model of the service before any agent is configured. Its notation follows the service-blueprint tradition — customer actions, onstage and backstage work separated by a line of visibility — combined with a multi-actor journey timeline, and the resulting map feeds the agent setup directly: onstage steps become agent scope, risky steps become action guards, marked transfer points become warm handoffs.

Sources: Salesforce Agentforce observability · Sierra supervisory agents · Parloa · the Magentic interface project, Microsoft Research 2025 · appropriate-reliance study, 2025 (n = 308) · service blueprinting after Shostack, Harvard Business Review 1984. All conversations here are scripted.