Last updated:
Quick answer
Once a week, export or copy what the agent did, from sent drafts and bookings to edits and errors, and have an AI sort it into done well, needed fixing and should not have happened. Tighten one rule for every problem.
Free PDF
Get this setup as a PDF
Confirm your email and we'll open a print-ready copy of this setup you can save as a PDF. You also get the free starter pack and one setup a week. Unsubscribe anytime.
Already a subscriber? Unlock it on this device
At a glance
All outcome figures are estimates from cited sources, not independent verification by us.
- What it is allowed to do
- Read-only
Only reads, researches, or summarizes. It never drafts outbound content or changes anything.
- Time saved (estimate)
- About 15 minutes a week.
- Difficulty
- Intermediate
- Tools needed
- ChatGPT, Claude
- Typical cost
- Free tiers can run the review. Prices: Not stated in sources.
Reviewed Nov 2, 2026
Basis (estimate): Estimate only: based on the steps above, not timed.
What does this setup do?
Spend 15 minutes a week reviewing what your AI agent did: sort its actions into fine, fixed and should-not-have-happened, then tighten one rule for each problem.
Who is it for?
Anyone running an AI agent on email, calendar or a store who wants to stop babysitting it daily without losing control.
How do you set it up?
Tools
- ChatGPT or Claude
- The agent's activity: sent folder, drafts, calendar changes or its own log
Steps
- Pick a fixed 15 minutes each week.
- Gather what the agent touched: sent folder, drafts, calendar changes, order notes or the agent's activity history.
- Remove personal details you do not need, then paste it into the prompt below with your rulebook.
- Read the should-not-have-happened list first. OpenAI says a signed-in agent can access sensitive data and perform actions, so treat these seriously.
- For each problem, change one thing: tighten a rule, remove an app connection, or move the action behind approval.
- Keep the weekly reports. Fewer problems over time is how you know you can trust it with more.
Example output
Illustrative. Your results will differ.
- Fine: 23 drafts approved and sent
- Needed fixing: 2 replies quoted last year's price
- Should not have happened: 1 calendar invite sent without approval (broke rule 3)
- Rule change: move calendar invites behind SEND approval
Where to paste this prompt
ChatGPT → Project instructions, or ChatGPT Tasks for scheduled reminders (once a day on Free)
Copy-paste prompt
Adapt bracketed fields to your situation.
Review what my AI agent did this week. Use only what I paste.
My rulebook: [PASTE RULES]
Agent activity this week: [PASTE ACTIONS, DRAFTS, SENT ITEMS, CHANGES]
Sort every action into:
1. Fine
2. Needed fixing (what was wrong)
3. Should not have happened (which rule it broke)
Then give me:
- The one rule change that would have prevented the most problems
- Anything that looks like it acted outside its permissions
Keep it under 300 words.Newsletter
Get the free starter pack
Free starter pack with five beginner copy-paste prompts after you confirm, plus one researched setup in your inbox each week. Unsubscribe anytime.
What usually goes wrong?
Where agents tend to stall on this job. More patterns in Where AI gets stuck.
- Reviewing only what the agent reports about itself; check the real sent folder too.
- Fixing problems in a chat instead of in the agent's standing rules.
- Skipping weeks once things look fine.
Which chat tools work with this prompt?
Paste the prompt into any of these assistants (free or paid tiers).
- ChatGPT
- Claude
Common questions
- How do I keep an AI agent in check without watching it all day?
- Review its actions once a week, sort them into fine, fixed and should not have happened, and tighten one rule per problem.
- Where do I find what the agent did?
- Your sent folder, drafts, calendar history, store notes, or the agent tool's own activity history.
- When can I give it more freedom?
- When weekly reviews show few problems for several weeks in a row, and only one new permission at a time.
Where should you go next on this site?
What tips make it smoother?
- Same day, same time each week makes it stick.
- Read problems first, wins second.
- If the same problem shows up two weeks running, move that action behind approval.
Where do the claims come from?
Summaries use public sources linked below, not independent verification. How setups are picked.
- Read OpenAI’s write-up (help.openai.com)
OpenAI: ChatGPT agent: when signed into sites or apps it can access sensitive data like emails, files or account settings and perform actions (read 2026-11-02 from a search result excerpt of the page)
- Read Anthropic’s write-up (support.claude.com)
Anthropic: Use Google Workspace connectors: Gmail drafting, approval before send, reply or forward by default (read 2026-10-31 through a page fetch tool, not a direct browser open)
Related setups
- Set guardrails before an AI agent gets access
Write a one-page rulebook for an AI agent before it touches your accounts: what it can read, what it can only draft, what always needs your approval, and what it must never do.
- Make your AI agent stop for approval
Make your AI agent stop and show you every email, booking or post as a draft with a one-line summary, and wait for your OK before it goes out.
- Pick your first AI agent job
Score your weekly tasks for time, repetition and risk, and pick one safe first job for an AI agent, with a 30-day test plan and a clear way to measure if it helped.