Skip to main content
The Marketing Agent coordinates the conversation. Specialists contribute focused work such as research, campaign planning, creative direction, or creative generation. The Agents workbench makes that coordination visible without forcing you to reconstruct it from chat messages. Use the workbench to answer three questions: who owns the current step, what did that specialist produce, and what still blocks the overall result?

Choose how the request is routed

The agent control in the composer supports two paths:
  • Keep Auto — Routes each request selected when you want the Marketing Agent to choose a perspective for each turn.
  • Select a specialist shown in the menu when you want to keep that specialist’s perspective for the conversation.
The available specialist names come from the current workspace. Selecting one changes the conversational perspective. A longer run still starts only when Marketer creates the task and, when shown, you select Approve & Run. For multi-step work, describe the result rather than trying to assign every internal handoff yourself:

Review the run before it starts

When Marketer proposes coordinated work, the proposal lists the steps it intends to run.
Marketer conversation with a seven-step Creative Pipeline proposal and Dismiss, Edit, and Approve and Run controls

The plan exposes the work before specialists or a pipeline begin the longer run.

Read the plan for evidence collection, production, ranking, and the final deliverable. Select:
  • Dismiss to stop the proposal.
  • Edit to change the goal, steps, inputs, output count, or review boundary.
  • Approve & Run to start the plan as shown.
This approval starts the listed work. It does not grant blanket authority for later changes to connected accounts.

Open the Agents workbench

After work starts, select the Open Agents control on the right edge of the conversation. The compact sidebench shows active and recent agent work. Select More to open the full overview.
Marketer Agents workbench with Active and Done sections listing campaign, research, and creative specialists with status, description, and update time

The full Agents overview separates current work from completed work and keeps each row tied to its run state.

Read the overview from top to bottom: An orchestrated family can contain a coordinating run, sibling specialists, and nested specialist work. The overview groups related runs so you can trace the family without assuming every row is independent.

Inspect one specialist

Select a row in the overview. The header changes to a breadcrumb beginning with Agents, followed by the selected specialist. The detail body shows the current or final run content. For a coordinated plan, the sidebench can expose:
  • Tasks: the plan steps and whether each is pending, active, done, or blocked.
  • Outputs: artifacts or result links produced by the run family.
  • Sources: the linked evidence made available for inspection.
Use the back control to return to the overview. Use the open-detail icon when you want the full Activity detail page for that run.

Follow a specialist into nested work

A specialist can open a nested pipeline or result. Select the specialist row, then select the nested work from its detail instead of stopping at the parent status.
Marketer workbench showing Agents, Campaign manager, and Creative Pipeline breadcrumbs, a running-time label, Stop control, three completed pipeline steps, and a result summary

The breadcrumb preserves the path from Agents to Campaign manager to the nested Creative Pipeline.

Use this review path:
  1. Confirm every breadcrumb level matches the intended agent family and nested run.
  2. Compare the top-level run state with the nested step states; completed nested steps do not necessarily mean the parent has stopped working.
  3. Read the result summary and open any linked output, not only the step checks.
  4. Check product identity, claims, brand direction, format, and review criteria.
  5. Follow any Outputs link to Studio or Assets and confirm the item exists there.
  6. Return to the conversation when you want to revise the brief or request a new direction.
The specialist result explains what its branch produced. The conversation remains the place where the Marketing Agent combines branches into the requested final answer.

Stop work that is no longer needed

When Stop is visible in a running specialist or pipeline detail, select it only when the objective should no longer continue. Marketer requests cancellation for that run and its child work, and the control changes to Stopping… while the request settles. Stay on the same detail until the run shows Canceled. Do not start replacement work while it is still stopping, because active child work may not have reached its terminal state yet. If Stop is absent, the run may already be terminal or your workspace role may not allow changes.

Handle each specialist state

Working

Open the row to see the active goal or step. A specialist can remain active while a sibling is already done. Wait for the combined result unless the active row requests input.

Waiting

Read the detail to identify what is missing. If it needs a fact or constraint, answer in the originating conversation. If it needs a recorded decision, open Awaiting action.

Completed

Open the output and sources. A completed research branch may provide evidence; a completed creative branch may provide an artifact. Neither is automatically the entire requested result.

Partially completed

Determine which branch produced usable work and which branch did not. Preserve supported output and ask the Marketing Agent to reconcile the missing part.

Failed

Open the failed row before retrying. Check the last successful step, error, required connection, and whether sibling outputs are still usable. Retry the failed branch only when the cause is resolved.

Canceled

Treat the run as stopped. Start a revised run from the conversation if the deliverable is still needed; do not assume a canceled branch will resume.

Reconcile multiple specialist results

When specialists return different recommendations, bring them back to one decision frame in the conversation:
Then verify the combined result:
  • Every conclusion maps to an output or source you can inspect.
  • Conflicting measurement frames are made explicit.
  • Generated claims match approved product and brand context.
  • A recommendation is not presented as an applied change.
  • The final result answers the original deliverable, not only one specialist’s branch.
Specialist identity and completion state do not prove a claim. Inspect the source or output behind material recommendations, especially those affecting spend or customer-facing creative.

Next step

Track the parent task

Connect specialist state to the task’s current step and terminal result.

Review pending decisions

Resolve an approval that pauses a specialist or coordinated run.

Review creative output

Continue from the run output into Briefs, Studio, and Assets.

Reopen the conversation

Return to the original request, attachments, and decisions.
Last modified on August 9, 2026