AI Prompts That Turn One-off Tasks Into Standing Agents

You have now seen my business: the connected tools, the files, the patterns in my email, where the money comes from, and where my attention goes.

Based on that evidence, design my agent roster: the five recurring agents that would save me the most hours or recover the most money.

Rank them by leverage.

For each agent, show:

1. Agent name
2. What it watches (which inbox, tool, or folder)
3. What it produces (drafts, a report, a ranked list)
4. How often it should run
5. The specific evidence from my business that justifies it
6. What could go wrong if it misfires
7. Risk level: low, medium, or high

Rules:

- Ground every agent in evidence you actually saw. Do not propose an agent for something that happens twice a year.
- Every agent starts draft-only: it prepares, I approve.

Then build the number one agent right now: write its standing instructions, set its schedule, and run it once in front of me so I can see exactly what will be waiting for me each morning.

Take this task and turn it into an agent that runs on a schedule without me starting it: [name the prompt or describe the job].

Set it up completely:

1. Write the standing instructions, including my voice and my red lines
2. Choose the schedule based on how often the underlying thing actually changes: daily for inboxes, weekly for money, monthly for pricing
3. Decide where the output lands so I see it without hunting for it
4. Start it in draft-only mode: everything it produces waits for my approval

Then define its promotion path:

1. Track its accuracy as I approve or correct its output
2. After it has been right twenty times in a row on some category of item, propose moving just that category to automatic
3. Bring me the evidence when you propose the promotion, and I decide
4. Promotions are carve-outs, not blank checks: record each one in the memory file with its exact limits, and my permanent red lines (committing me to money or legal terms) are never up for promotion

Run it once right now as a test so I can see the output before it goes on the schedule.

Build me an agent that makes sure nothing with money attached ever slips again.

Every morning it sweeps my email and open proposals for:

1. New inquiries that arrived since yesterday
2. Proposals with no reply after five business days
3. Anything I promised in writing that shows no evidence of being delivered

For each item it finds, it drafts the right response in my voice from my sent mail, all marked HOLD.

Then it hands me one short morning list:

1. What needs a decision
2. The drafts, ready to approve
3. Nothing else

If a morning is clean, the report is one line saying so.

Build it, run today's sweep right now, and show me the output before you put it on the schedule.

Build me an agent that runs before I start work every Monday.

It produces one page:

1. Current receivables aging from my connected accounting tool
2. Payments that arrived last week
3. Chase drafts for anything genuinely overdue, cross-checked against payments received so it never chases someone who already paid
4. The one number I should be watching this week, with the reason in plain English

All drafts HOLD.

Build it and run the first one now, so Monday's version arrives in a format I have already approved.

Look at the evidence of my business and find the place where I or my clients keep doing something manually that a simple tool would fix.

Look for patterns like:

- Quotes that take an hour to assemble from old proposals
- An intake process that lives in six back-and-forth emails
- A calculation I redo in my head every time someone calls
- A status question clients keep emailing me to ask

Propose the one tool worth building first. Describe it in plain terms:

1. What I would see
2. What my client would see
3. What manual work it replaces
4. What it would need to stay accurate

Then build a working version I can try today.

Rules:

1. It runs on my machine only. Nothing public.
2. It touches no live client data until I say so.
3. Take the time to check your own work before you show me. I care that it works, not that it was fast.

When I have used it and it holds up, we will talk about where it should live permanently.