Playbooks
How to Run AI Agents on a Schedule With Routines
Run AI agents once, at an interval, or on set weekdays. How Brainwrite routines, results threads and run receipts work, and when your Mac must be on.
To run an AI agent on a schedule, give the agent a saved prompt and a timing rule, then let the app start it for you. In Brainwrite this is a routine. A routine runs once, at a repeating interval, or on chosen weekdays. Each run starts fresh, can post its report to one results thread, and leaves a receipt.
What is an AI agent routine?
A routine is a prompt you would otherwise type every morning, saved with a bot and a schedule. When it falls due, Brainwrite starts a new run using that bot's current model, permissions, tools and computer setup.
Routines suit work that repeats at a known time:
- A morning brief from your calendar and inbox
- A weekly report pulled from your project tracker
- A daily competitor watch that checks pages for changes
- A weekday review of sales threads that went quiet
If the trigger is an event rather than a time, such as a form submission or a deploy, use a webhook instead. Webhooks start the same task executor from an HTTP request.
How do you schedule an AI agent in Brainwrite?
You can create a routine from a bot's settings, or ask the bot to set one up in chat. The steps in the editor:
- Pick the bot. The routine runs as that bot, with its access and model.
- Write the prompt. Be specific about scope and output. Name the mailbox, the date window and the format you want back.
- Choose the timing. Once, a repeating interval, or selected weekdays at a set time.
- Choose where results go. Use Post results to to pick an existing thread, or Create a dedicated results thread.
- Save, then check the first run. Read the first report before you trust the routine.
Write prompts that survive a fresh start
Each run starts a fresh execution. It does not carry the previous run's conversation. So the prompt has to stand on its own. A good routine prompt says what to read, what to skip, what to return, and what not to do.
Here is the routine that ships with the Recover Unanswered Sales Emails team in the team library, which runs on weekdays at 08:30:
Review the approved sales mailbox and optional CRM scope for conversations where our team owes the next step. Exclude opted-out, closed, duplicate, automated, internal, and low-confidence threads. Return a prioritized review queue with owner, last meaningful exchange, promised next step, age, confidence, and a draft request for Nudge. Do not send or modify records.
Notice the last line. A scheduled run should usually prepare work for you to approve, not finish it on its own.
Where do the results of a scheduled AI agent go?
Every run gets its own thread, titled with the routine's name and start time, such as Daily briefing · Oct 3, 9:00 AM. Left alone, a daily routine would fill your sidebar. Results threads fix that.
Keep reports in one results thread
With Post results to, each run still executes in its own context, but its dated report lands in the thread you chose. The background execution threads stay out of the normal sidebar. You can move the results thread into a folder, such as Reports, like any other thread.
From a report, Open run shows that run's full conversation and any approvals. Back to results takes you back to the list.
A few rules worth knowing:
- Routines created in chat keep reporting to that conversation unless you pick another destination.
- Changing the destination affects future runs. Links on older reports stay as they were.
- If you delete a results thread, its hidden execution threads become visible, so you can still inspect their work and answer approvals.
How do you know a scheduled run actually worked?
Every run leaves a receipt with its status. Run logs keeps the full history.
| Status | What it tells you |
|---|---|
| Queued | The run is lined up to start |
| Active | The run is working now |
| Waiting | The run needs something from you, such as an approval |
| Completed | The run finished |
| Missed | The run did not take place |
| Failed | The run hit an error |
| Cancelled | The run was stopped |
Check the receipts in the first week of any new routine. A run that keeps waiting usually means the prompt asks for an action the bot is not allowed to take on its own. That is the system working. Either approve it each time or rewrite the prompt so the bot prepares the work instead.
Does the app need to be running for scheduled AI agents?
On your Mac, yes. This is the edge to plan around.
Brainwrite must be running when a local routine becomes due. Local routines and webhook deliveries are not hosted jobs: closing the desktop app stops the local executor and the webhook receiver. Your Mac also has to be awake.
You have two options:
| Local routines | Brainwrite Cloud | |
|---|---|---|
| Where runs happen | Your Mac | Your own always-on Cloud computer |
| Needs your Mac on | Yes, awake with Brainwrite open | No |
| Good for | Work during your day, local files and apps | Overnight and early-morning runs, all-day jobs |
| Plan | Included with the desktop app | Brainwrite Cloud, $99 a month |
Brainwrite Cloud runs your bots around the clock, including routines and connected apps, even when your Mac is asleep or switched off. It is a Linux computer that belongs to your account alone. AI usage is not included. You sign in with your own Claude or ChatGPT account, or add an API key, on the Cloud computer as you do on your Mac. For bots that run all day, the Cloud docs recommend a Claude Max plan or an API key.
If a job needs something only your Mac has, such as an app or a signed-in site, you can lend your Mac to your Cloud while it is awake and Brainwrite is open on it.
Why do team library routines start paused?
When you add a team from the library, its routines install paused. Skills arrive on, but connected apps stay off until you approve them. A routine from someone else's template should not start reading your inbox the moment it lands.
To turn one on:
- Read the routine's prompt and schedule.
- Grant the bot only the apps the routine needs. See connected apps.
- Run it once by hand and read the result.
- Unpause it.
Common mistakes when scheduling AI agents
- Vague prompts. "Check my email" gives a different answer every day. Name the scope.
- Actions without a human. Have scheduled runs draft, not send.
- No results thread. Without one, reports scatter across the sidebar.
- Assuming the Mac is on. A laptop asleep at 7:00 means no 7:00 brief. Use Cloud or schedule for when you are working.
- Ignoring cost. Each run uses tokens on your provider account. Check per-thread usage and cost after the first week.
The routines documentation covers every option in the editor.
Try it
Brainwrite runs on macOS today. Download the app and choose a plan on the pricing page. Take one prompt you already run by hand, save it as a routine with a results thread, and check its receipts for a week before you add a second one.
