Loading
In this lessonBy the end, you’ll be able to
  • Explain how connections, corrections, untagged replies, and routines add up to ownership
  • Shift from directing tasks to setting goals and reviewing the decisions that matter
  • Hand Claude one ongoing job, with clear lines on what comes back to you

What these features add up to

Throughout this course, you have been deepening how Claude understands your work.

Together, these give Claude your team's context, let your colleagues shape its work, and keep it working toward a goal for weeks, not only for one request.

These features are visible and can be managed from the channel's Configure page, which opens from the Configure link under any Claude reply. Each channel has its own. Any full member of the channel who is signed in to your organization's Claude account can open it; what you can change there depends on how your organization has set it up. Click through it here:

Loading

From tasks to an ongoing job

Claude can run independently across multiple channels, doing work that used to require repeated prompts from you. With Claude in several channels and several tasks running at once, you are able to delegate and steer work at the level of a team rather than one task at a time.

Loading

Over time, as you verify more of Claude's work and start to trust what it produces, you can scope it to larger pieces or projects that involve more continuous work and decision-making.

Your role shifts from task-by-task direction to setting goals and directing larger projects. You set the standard once, Claude handles the routine cases, and you review only the decisions that matter. Your attention is the limit now, so ask Claude to group its questions and to bring each decision with what it recommends. This is what teams at Anthropic do: delegate work in bounded ways, keep human judgment for the calls that need it, and improve the system as Claude learns what works (Building effective human-agent teams).

Loading

Try it

Hand over one ongoing job

In a channel for this work · ongoing; first check after a few days

Pick something you keep healthy over time, like a launch, a stream of decisions, or a resource people depend on. For a launch, that means Claude watches the tracker and the channels, works out why something is stuck, finds what is missing, and brings you the decisions only you can make.

  1. Give the work a channel: create or pick one where this work will live; you and Claude both post there.

  2. Connect what Claude needs to read: the tracker, the channels where the work happens, the docs where decisions land (lesson 3). Ask what can you access from here? to confirm.

  3. Tell Claude the goal, what it owns, and which decisions to check with you on: the goal is the channel's north star, one measurable outcome that a person sets. With a goal to work toward, Claude can propose work, not only respond when something is stuck. Include something like this in your message to Claude:

    "The goal: we launch on October 14 with no open blockers. You own keeping us unblocked until then. Read the tracker and these channels. When something is stuck, find out whether we need information or a decision. Get the information yourself. Bring me anything about scope, timeline, or a real trade-off, grouped twice a week."

  4. Correct Claude as it goes: when it surfaces something useful or misses something, say so in the thread and save the pattern (lesson 6).

Done when: the ownership message is posted in the channel, Claude has acted on it once without being asked, and it has brought you one decision instead of making it.

Before you move on

Key takeaway: extend trust a step at a time, per kind of task: from tasks you check to an ongoing job Claude runs, with clear lines on which decisions come back to you.

After practicing: You will have written down one ongoing responsibility for Claude in a channel it reads, including what it decides and what it brings to you.