Skip to content

Agent

Work with the Vessel agent

Describe the outcome, provide context, and review the work.

What you’ll work toward

Describe the outcome, provide context, and review the work.

Ask for a specific outcome

To use a new recording, choose Attach recording and wait for it to be ready, or paste a YouTube link in your message. Tell the agent what you want made from it and who it is for. Preparing a transcript may use points; review any cost and approval shown before continuing.

Open the agent workspace and describe what you want to accomplish. Give the sermon or project, audience, format and constraints. For example: “Help me turn the forgiveness sermon into a short reflection for small-group leaders; preserve the qualification about reconciliation.”

The agent can help navigate and work with supported tools. Follow any action or navigation control it provides, and inspect the resulting project in its own workspace. If the agent asks for missing context, supply the relevant source or clarify the request.

Stay in control

Add any missing context in the message field, then send your request. Review the named action and its cost in an approval card before approving. Background preparation can finish after the chat pauses; watch the task status and open the saved result to review it.

Review generated text, scripture, media and destinations before sharing or sending. For longer work, return to the project or run status to see whether it finished. An agent response is a starting point for your judgment, particularly when content will represent your church publicly.

Give the agent a useful brief

In the sample Harbor workspace, a communications lead might ask: “Use the ‘Practicing Patience’ sermon to help me prepare a short reflection for group leaders. Keep the distinction between patience and passivity, and suggest one discussion question.” This identifies a source, audience, outcome and constraint. If there are several similarly named sermons, specify the title or select the intended project when prompted.

The agent may return a navigation or action control that opens the appropriate workspace. Follow that control and inspect the actual project, rather than assuming a chat summary is the final deliverable. An agent can help coordinate supported tools, but the options available depend on the task, source and your access. Give it additional context when it asks instead of repeating the same ambiguous request.

Review the result and its state

Read generated content against the sermon; verify scripture and any quotation. For a clip, check the crop, captions and exported sound. For an email, check recipients, links and delivery separately. A longer operation may continue after the initial request; open the project or run details to check whether it has completed or needs attention.

If the agent cannot find a source, confirm the active workspace, the exact title and whether the source has finished preparation. If it proposes a result that misses the audience, explain what needs to change and inspect the new output. Keep sensitive credentials out of the conversation.

Next, choose a tool, build an Action, or review a transcript.

Break a larger request into checkpoints

For a multi-step project, ask the agent to identify the source and propose a direction first. Check that it selected the intended sermon, then ask it to prepare the draft or open the relevant workspace. This is particularly useful when two recordings have similar titles. A specific correction—“keep the speaker’s qualification in the second paragraph”—gives the next pass more direction than “make it better.”

Before any delivery, open the actual output and confirm its state, destination and audience. The agent’s conversational description and the file a reader receives should agree.

Was this a useful next step?

Your next step

Take the idea into the workspace.

Keep this guide nearby. Start small, review the result, and make it your own.

Try it in Vessel