Set up a daily
AI business brief.
Give ChatGPT desktop, with Codex, one business folder and a short assignment: read the current notes, find what needs attention and save a morning brief. Check it once, then schedule the same job.
Published · Updated
You open your computer and spend the first part of the day remembering what you were doing. Which draft needs a correction? What are you waiting on? What can move today?
A morning brief is a good first recurring AI task because you can check it against the notes you already have. Start with a few approved files. You can add connected tools when the job needs them.
Set the schedule · Get guided setup · $20
Can I run this with only a phone?
No. This guide runs a local job in Codex on your computer. The computer must be on, the desktop app running and the project available when the job runs. Your phone can help you prepare notes and review guidance; it does not host this local shift.
The $20 kit does not include a Fort Rock cloud runner or automatic delivery of the brief to your phone. A separate cloud workflow needs its own supported setup. Your computer still has to show up for its shift.
What should a daily AI brief include?
- Needs your answer: missing facts, conflicting instructions and decisions blocking work.
- Ready to review: saved drafts, with a link to each file.
- Next useful action: a short recommendation supported by the notes.
- Sources: which files it read and anything it could not check.
Keep the first version short enough to read. An empty work log should produce “no recorded update,” rather than an invented story about yesterday.
Try a brief with three practice notes.
Here is a fictional workshop example you can check in a few minutes. Put these three notes in a practice folder:
business.md
We run Saturday pottery workshops. Do not promise a place until attendance is confirmed.
next-actions.md
The next workshop needs a confirmed start time. A reminder draft is ready to review.
work-log.md
No completed work has been recorded.Ask your agent:
Read business.md, next-actions.md and work-log.md. Save a short brief with the decision I need to make, the draft ready for review, the next useful action and the source for each point. Keep missing facts unknown. Do not send a reminder or create a schedule.
A useful practice result would say:
- Needs your answer: confirm the workshop start time. Source: next-actions.md.
- Ready to review: a reminder draft is mentioned, but no draft file was supplied. Ask where it is before claiming to have reviewed it.
- Next useful action: confirm the start time, then review the actual reminder. Do not promise attendance. Sources: next-actions.md and business.md.
- Completed work: no recorded update. Source: work-log.md.
Check whether it invents a time, claims to read the missing draft or quietly turns a draft into a sent message. Correct the source note, repeat the brief and see whether the correction carries through. Then try your own approved notes.
Make it read today’s facts.
A recurring prompt should describe the job. The folder should hold the changing facts. Save corrections there so tomorrow’s run can find them.
Use an approved business brief, your current next-action list and selected drafts. Keep their paths in the assignment. An email summary needs access to email; a folder of notes only tells the agent what is in those notes.
Schedule the brief after the first check.
In the same desktop project, ask for a local scheduled task using the assignment you tested. Choose your own time and time zone. For example:
Schedule this checked brief for weekdays at 8:30 a.m. America/Los_Angeles, in this local project. Each run must read the current source files again, follow the same review rules and save a new dated brief under work/briefs/ without overwriting earlier results. Keep it to drafts. Show me the saved task, project, time zone and next run.
Open Scheduled and check those details. A saved schedule is the starting point. After its first run, open the resulting file, check the sources and confirm that the work actually finished. Pause a practice task when you’re done.
Local work needs the computer on, the desktop app running and the project available. Account features and limits vary. OpenAI’s scheduled-task documentation explains local projects, run history and controls.
What we checked ourselves.
We tested a one-time morning shift against a separate copy of a kit workspace. The scheduler started it, the agent saved and reopened a source-linked brief, and we checked the result before pausing the task. The work log was empty, so the brief left business progress unknown. That proves the small workflow ran in our test; its usefulness for your business depends on the notes and job you give it.
Four more jobs for the same habit.
Once the brief works, try a weekly review, content drafts, lead follow-ups or an inbox sort. Each needs its own sources, output location and review check. Start with one shift you will actually read.