When to use Agent Workspace
Use Agent Workspace when a task needs multiple steps, clear progress states, reusable specialists, or better visibility into what is running and what was produced.
A simple one-step question may not need a run. Use the lightest workflow that still keeps the work understandable.
Create a run
Choose the active project
The run should start with the correct project context.
Define the task
State the goal, expected result, and limits.
Choose the specialist or launcher
Select the configured capability that fits the task.
Review context
Confirm the files, instructions, and decisions included.
Launch
Start the run and watch its state.
Understand run states
Queued
Waiting to begin.
Active
Currently running.
Needs attention
Waiting for a decision, input, or correction.
Completed
Finished with a result ready for review.
Failed
Stopped before a usable result was produced.
Review the result
Inspect the output, verify it against the requested goal, and decide whether to capture it, revise it, or run another step.
Return the result to the project
Artifact
Save a useful deliverable or reusable result.
Task update
Mark progress or create the next action.
Decision
Record an important choice and why it was made.
Handoff
Prepare the next person, model, or session to continue.
Keep runs understandable
Use clear run names, keep the task narrow, avoid sending unrelated project material, and close or archive stale runs.
Ask LobeWork about this guide.
The question will be placed into the support widget so you can continue from the exact topic.