An agent combines a prompt with a trigger. Use one to prepare a weekly portfolio review, research renewals, or turn a newly received meeting transcript into follow-up work.

Create an agent

Open Agents and create an agent in My Agents, or add a template from Workspace or Discover. Give it a name and a prompt describing the data to use and the result you want. Type / in the prompt to include a skill. Tags help you organize and find agents.

Choose a Trigger:

Trigger	When it runs
On a schedule	At the time and recurrence you choose. Check the timezone shown with the schedule.
Manually	When you start a run yourself. Saving it does not schedule recurring work.
When a meeting transcript is received	Once per eligible meeting you attended, with the meeting and company attached as context.

Run and verify your first agent

Start with a result you can use in your next portfolio review. Create an agent named Weekly customer decisions, choose Manually for the first run, and use a prompt such as:

Review Pfizer, Cloudflare, and Siemens using their latest customer conversations, open work, goals, and customer commitments. Prepare a decision brief for our CS leadership meeting. For each account, identify the outcome at stake, decision owner, deadline, what changed since the previous week, and the evidence still missing. Link the original sources and include their dates. Separate confirmed results from estimates. Recommend one next action with a proposed owner and date. Finish with the decisions our team needs to make this week. Keep the brief in this chat.

Then complete the run:

Save the agent and choose Run now. Scheduled agents also offer Run now, so you can check their output before the first scheduled run.
Open the new run from its history. If a run is already in progress, open that existing run to see what it is doing.
Follow any Waiting for your input indicator into the conversation. Review the proposed operation and approve or decline it using Assistant approvals.
Read the finished output. Open its source references and check the owners, dates, and unresolved decisions against the original records.
Turn the proposed next action into work. For example, assign Alex Morgan a task to obtain Pfizer's Payroll evidence. Keep Grace Chen's customer commitment in Waiting on them, and keep renewal approval open until Finance decides. If you ask the agent to create the task, open it and verify its account, teammate assignee, and due date.
When the output meets your team's needs, edit the agent to use On a schedule. Set the cadence and timezone for your leadership meeting, and check the displayed Next run.

Start on Home, open the reusable agent, and inspect its completed review. Check the linked evidence before turning the proposed next action into assigned work.

The completed review brings each account's decision date, supporting sources, missing evidence, and next action into one brief.

Use the brief to decide where the team needs to act this week. In the example scenarios, that means distinguishing Pfizer's verified rework improvement from its pending Payroll evidence, Cloudflare's three completed pilot outcomes from its blocked fourth pilot, and Siemens' closed routing fix from the eight acceptance cases still outstanding.

Meeting transcript agents

Choose When a meeting transcript is received and write a prompt to review decisions, commitments, and proposed follow-up tasks. Save the agent when its configuration is ready.

Connect your calendar and a call integration. The transcript must be linked to a calendar meeting, and your workspace user must be an attendee who has not declined it.

The trigger is for recent meetings, normally those that started within the last 48 hours. An initial import of old calls does not trigger a backlog of runs. Re-syncing a transcript does not start the same agent again for the same meeting. A paused agent does not react to incoming transcripts.

For example, ask the agent to summarize decisions, identify commitments, and propose follow-up tasks. Review its output and any requested approvals before relying on the results.

To verify this trigger, keep the agent active and let a new eligible meeting transcript arrive. Open the resulting run and check that its attached meeting and company match the call. Transcript agents run from that event and do not offer Run now. If no run appears, check the calendar link, your attendee status, the meeting's age, and whether this agent has already processed that meeting.

Templates and ownership

Workspace templates give teammates a common starting point; templates do not run on their own. Add one to My Agents to create your own agent. Builders, admins, and owners can manage workspace templates.

For an agent linked to a template, Keep this agent up to date with its template applies template changes automatically. Turn it off to customize the agent's settings independently.

Review results and manage runs

Open an agent to inspect its runs and results. Opening a result marks it as reviewed. You can also use the Assistant's sessions list to return to background work.

Run indicator	What to do
Running in the background	Open the run to follow its progress. Use Stop in the conversation if you need to interrupt it.
Waiting for your input	Open the run and answer the requested approval. It is waiting for your decision.
Failed on last run or an error message	Read the error and any completed work. Resolve the cause, then use Retry in the conversation to continue from its saved context.
Not viewed yet	Open and read the result. This indicator tracks whether you have opened it, separately from whether the run finished successfully.

For retry and stop behavior, see Sessions, stop, and retry.

By default, Cust can suspend an agent when its two most recent runs are both unreviewed. Open the agent to resume it or adjust its settings. Keep running even if I don't review results opts out of that suspension; those runs continue consuming credits.

Pause an agent to stop its automatic triggers. Resume it when you want future runs to continue.

Open Agents