Understand the signed-in agent surface across copilot threads, scheduled automations, delivery routing, and workflow-console operations.
Use this guide when you need to understand how the Agent Workbench combines persistent copilot threads, scheduled automations, delivery, and workflow orchestration into one workspace-bound surface.
The current Agent Workbench page is organized as a sequence of operational surfaces rather than separate routes.
Use persistent workspace-bound threads with provenance, citations, and follow-up history.
Create recurring briefs, alerts, drift summaries, and promotion-gate checks with profile-aware delivery.
Configure where outputs land, then triage automation and copilot results from a shared inbox surface.
Audit copilot threads, automation runs, tool calls, delivery events, and policy actions in one explainability rail.
Run structured MCP-native workflows and review orchestration history plus guardrail detail.
Use this sequence to keep agent workflows persistent, reviewable, and policy-aware.
Use a workspace-bound thread to frame the task, preserve provenance, and keep follow-up context stable.
Once a thread turns into a recurring job, formalize it as an automation instead of re-running it ad hoc.
Set delivery and inbox behavior so the right workspace or user sees the output without losing traceability.
Use the timeline and workflow console to confirm tool use, policy events, and orchestration history before trusting the result.
The first half of the workbench is where research intent becomes scheduled, routed work.
Start from workspace-bound thread history so follow-up prompts inherit prior answers, provenance, and citations instead of restarting the context each time.
Use the automation center to turn stable prompts into recurring jobs with workspace-aware cadence and profile assumptions.
Configure where agent outputs land before you scale a workflow. Delivery routing is part of the workflow contract, not an afterthought.
The delivery inbox is where generated output gets triaged, acknowledged, and handed into research or execution workflows.
The second half of the workbench is where you verify what the agent actually did.
Use the timeline to see threads, automation runs, tool calls, and delivery events in one place before trusting the result.
The workflow console is the structured execution surface for MCP-native workflows. It is where run history and guardrail detail stay visible.
Check policy and tool events when output quality changes unexpectedly. They explain whether the issue came from permissions, routing, or execution.
Keep the trace from prompt to delivery intact so downstream teams can see how an agent result was produced.
Use these scenarios when a thread or automation needs a concrete operator playbook instead of a generic walkthrough.
Start from the thread that already has the right workspace context, convert it into an automation with a named cadence, and route the result into a delivery channel the desk actually reviews each morning.
The recurring brief inherits the original thread context and arrives in a predictable place without re-prompting from scratch.
Use the activity timeline and workflow console together to locate the failed run, inspect the policy and tool events around it, and confirm whether the problem came from permissions, MCP availability, or prompt drift.
You end with a specific failure source and a traceable record before rerunning or escalating the automation.
Open the inbox item, match it back to the originating thread or automation, then confirm the delivery channel, timeline entries, and workflow-console history all point to the same run.
The downstream consumer gets an output with intact provenance instead of an orphaned summary.
Jump into the signed-in agent surface and review copilot, automations, delivery, and workflow history.
Cross-check the MCP tools, resources, and bridge contracts that agent workflows rely on.
Review the runtime and agent-profile controls that shape automation and copilot behavior.
Mar 24, 2026
Report unclear guidance, stale contracts, missing coverage, or broken docs UI on this page.
Open feedback issueJump to the section you need without losing your place.
Mar 24, 2026
Report unclear guidance, stale contracts, missing coverage, or broken docs UI on this page.
Open feedback issue