Use d5s

Use d5s when the work benefits from a conversation, durable context, and a visible record of what the agent did. You can begin with a chat; you do not need to configure an agent or automation first.

Start with one concrete outcome

From the workspace home, start a chat and describe the result you need, the evidence the agent should use, and any boundary it must respect. A useful request is specific enough to review:

Review these interview notes, group the recurring objections, and prepare a one-page brief. Cite the note behind every conclusion and call out uncertainty.

Choose an existing project when the task should use its standing instructions, files, memory, skills, or connector activations. Otherwise, start directly; d5s creates an unlisted project behind the chat.

Follow the work, not just the answer

During a run, the conversation can show progress, tool calls, files, sub-agent activity, connector actions, and approval requests. Check that the agent used the intended sources and that any external action matches your request before approving it.

Continue in the same chat while the goal is changing. Ask the agent to correct assumptions, show missing evidence, or revise the deliverable instead of starting over.

Keep useful context

Use Save as project when the chat has accumulated instructions or material that should be findable and reusable. This promotes the project already behind the chat; it does not copy the conversation into a new container.

When the pattern stabilizes, choose what should happen next:

• keep it as a chat when a person should steer each step;
• staff an agent when a persistent role should own it;
• create an automation when the same job should run independently; or
• use a dashboard when people should return to a maintained output.