Skip to content
brazosd
User guideTHE FIELD GUIDE

Start with a question

Search every guideEsc
Browse guides
User guideEveryday work2 min read

Scheduled work

Ask an assistant to run a task once at a specified time or repeat it on a regular schedule. Test the job manually before automating it.

On this page

Choose a suitable task

Good candidates include periodic summaries of a known folder, weekly project reports, or a one-time reminder. A schedule uses the tools, materials, and permissions available when it runs; it does not create new access.

Create a schedule in conversation

Specify the task, time zone, frequency, input location, and where results should go.

Try this request
Create a task for every Monday at 9 a.m., using the Asia/Shanghai time zone.
Summarize the previous week's files in the organization's Project Updates folder.
Save the result in its Summaries folder. Do not send it externally.
After creating it, tell me its name, schedule, and next execution time.

For a single run, say “Remind me to review this file in 30 minutes, once only.” When using “tomorrow” or “next week”, ask the assistant to repeat the exact date and time zone.

Verify that it exists

A conversational promise is not proof of creation. Check the tool result and ask the assistant to list the task with its execution rule and state.

Test one execution before relying on a recurring job. Email, external messages, and notifications depend on configured channels and tools; do not assume a schedule sends a phone notification.

Pause, resume, or delete

Ask the assistant to list current schedules, pause a named task, resume it, or delete it. If names are ambiguous, review the list before choosing.

Pausing stops future scheduled work temporarily; deletion removes the rule. If an execution has already begun, inspect its session and stop that run separately if necessary.

Check dependencies

Inputs must remain accessible. A local computer or Chrome must be online if the task uses it. The organization needs sufficient balance and resources, and somebody may need to respond to authorization requests.

Automatic runs consume model and tool resources. After a failure, inspect the cause and completed actions before rerunning, especially if the task can send messages or change data.