Skill · 6 min read

The recurring AI task brief

AI tools are getting better at scheduled actions, projects, memory, and connected app context. The useful habit is turning a repeated prompt into a clear routine before you automate it.

A lot of AI productivity news is now about continuity. Google is rolling out scheduled actions in the Gemini app so recurring prompts can run at a chosen time. OpenAI's Projects guidance frames ChatGPT projects as dedicated spaces for long-running work with files, instructions, and shared context. OpenAI's app and connector guidance shows common workflows such as drafting executive updates from project information.

The practical lesson is not "schedule everything." It is to define the routine before you automate it. A recurring AI task should have a cadence, a source boundary, a fixed output shape, and a review rule. Otherwise you just get a surprise draft on a timer.

The skill

A recurring AI task brief is a one-page contract for work you want AI to help with repeatedly. Use it before setting up a scheduled action, project routine, weekly digest, automated research summary, or recurring content idea generator.

Recurring AI task brief

Routine:
{name the repeated task}

Cadence:
{daily, weekly, monthly, one-off reminder, or event-based}

Trigger:
{time, date, project milestone, inbox state, meeting, report cycle}

Allowed inputs:
- {documents, inboxes, apps, folders, dashboards, links, date ranges}

Excluded inputs:
- {private areas, old archives, draft folders, unrelated projects}

Output format:
{exact structure, length, tone, and sections}

Review rule:
{what needs human checking before sharing or acting}

Stop condition:
{when AI should say it cannot complete the routine reliably}

A worked example: Monday project pulse

Suppose you want AI to prepare a Monday morning project pulse. A vague recurring instruction is: "Every Monday, summarize my project." That sounds useful until it pulls from stale notes, misses blockers, or writes in a style nobody can scan.

The brief version is much safer:

Routine:
Monday project pulse for the launch team.

Cadence:
Every Monday at 8:30 AM.

Trigger:
Start of the work week.

Allowed inputs:
- Project folder: Launch Q3
- Meeting notes updated in the last 14 days
- Open action tracker
- Current risk log

Excluded inputs:
- Archived launch folders
- Personal notes
- Draft strategy files marked "do not share"

Output format:
1. Three-sentence executive summary
2. Completed last week
3. Blockers
4. Decisions needed
5. Suggested owner follow-ups

Review rule:
Do not send to the team. Prepare a draft for review.

Stop condition:
If the action tracker or risk log is missing, say what is missing and do not invent status.

The prompt

Use this before creating any recurring AI routine:

I want to turn this repeated prompt into a recurring AI routine.

Repeated task:
{what I want AI to do}

Audience:
{who will read or use the output}

Current source material:
{where the AI should look}

Create a recurring AI task brief with:
1. Routine name
2. Cadence
3. Trigger
4. Allowed inputs
5. Excluded inputs
6. Output format
7. Review rule
8. Stop condition
9. A first-run checklist

Be conservative. If the routine could send, publish, change records, expose private data, or make a recommendation based on stale information, require human review.

The first-run checklist

Why it works

Recurring AI is powerful because it saves the friction of remembering and re-prompting. It is risky for the same reason: if the routine is poorly scoped, it can keep producing confident but weak output on schedule.

The brief keeps the automation honest. It turns "do this every week" into a repeatable workflow with visible boundaries. That is the difference between a useful assistant and a noisy reminder machine.

Try it today. Pick one prompt you repeat every week. Do not schedule it yet. First write the cadence, allowed inputs, output format, review rule, and stop condition.

Sources

Keep reading

Related posts

Skill · 6 min read

The AI action checkpoint

Decide when AI may read, draft, stage, act, or stop for approval.

Read the skill →
Skill · 6 min read

The AI handoff note for long-running work

Carry context, decisions, constraints, and risks across AI sessions.

Read the skill →