Queued
Has the task started?
A recorded request is waiting to run.
status: queuedMarketingHQ · Personal agent workspace
Give delegated
work a shape.
A workspace with specialised agents that I used for complete tasks in my day-to-day work.
Inside the workspace.
Actual MarketingHQ interface · Demonstration data
The scheduling view
Source: the MarketingHQ Agent Activity page, running in an isolated local preview. Authentication and data adapters were replaced for the capture; the Agent Activity component was preserved.
Workspace
Context stays with the work.
Task → specialised role → run record → output
Strategist
Direction and planning
Researcher
Research and audience material
Copywriter
Written assets
Ads optimiser
Campaign optimisation tasks
Analyst
Analysis tasks
Developer
Implementation tasks
One run recordTask · state · timing · model · cost · output
What happens
after you delegate?
The interface includes run history and scheduling. The data model distinguishes queued, running, completed, failed and cancelled work.
Run states · Illustrative view
Running
What is happening now?
Active work is distinct from a pending request; the model records its start time.
status: runningCompleted
What came back?
An output is ready to inspect. Completion alone does not establish usefulness or correctness.
status: completedFailed
What needs attention?
Failure has its own state and error-message field, distinct from an empty successful result.
status: failedCancelled
Did this work stop?
Cancellation is represented separately from completion and failure.
status: cancelledMake the work
inspectable.
These are three concrete surfaces found in the implementation, and why they matter to the experience.
Interface
History & schedules
Agent Activity separates previous runs from scheduled work. A task can be located in time rather than disappearing into a conversation.
Run record
Input & output
task_type
input / output
status / error_message
started_at / completed_atSelected field names from the agent_runs schema; no private execution data.
Dependencies
Related work
depends_on
triggered_runs
retry_count / max_retriesThe model can represent relationships and retry counts. These fields alone do not prove that every recovery path works.
From everyday use
to a design question.
I used complete workflows in MarketingHQ. The recurring product question is how to provide useful context, understand what an agent is doing and review the result before acting on it.
Evidence and scope
This page combines my account of using the product with the local Agent Activity interface and agent_runs migration. The visualisations explain that structure; they do not claim audited reliability, measured savings or a current public deployment.