1. The Workflow Mapper
Turn a hand-drawn workflow into an automation map — start here.
I'm going to show you a workflow I currently do by hand. [Paste a description — or attach a photo of your sticky notes / whiteboard.] Interview me one question at a time until you fully understand every step, who does it, what tools are involved, and how long each step takes. Then give me back: (a) the workflow as a numbered list, (b) which steps AI can do today with tools I already have, (c) which steps genuinely need me, and (d) the single best 20% to automate first, with your reasoning.
2. The 20% Finder
Find your best first automation when you don't know where to start.
Interview me about my typical work week, one question at a time — my recurring tasks, what I dread, what I do late at night, what only I can do. Then rank my top 5 automation candidates by (time saved x how easy it is to automate) and recommend ONE to start with. Assume I'm using [Claude/ChatGPT] with no coding.
3. Build My First Agent
The 60-second scheduled-task agent, with guardrails built in.
Write me a reusable instruction block for a scheduled task in [Claude/ChatGPT]. The agent's job: [WHAT IT DOES, e.g., 'every Friday at 8 AM, collect this week's podcast transcripts from my Google Drive folder X, write show notes, a LinkedIn post, and a Substack draft, and save them back to folder Y']. The instructions must include: its role in one sentence; the exact steps in order; the output format; and this guardrail — if it hits a step it can't complete or information it doesn't have, it stops and tells me instead of guessing. Then tell me exactly where to paste this and how to schedule it.
4. The Agent Job Description
Write an AGENTS.md-style instruction file for the advanced path.
Turn the workflow below into an agent instruction file (AGENTS.md style). Sections: Role (one sentence), Tools it may use, Schedule/trigger, Step-by-step process, Memory (what it should remember between runs), Guardrails (what it must never do without asking me), and Escalation (when and how it hands things back to a human). Write it so another AI could run this job with zero extra context. Workflow: [PASTE].
5. Podcast Post-Production Agent
Transcript in — show notes, LinkedIn post, and newsletter draft out, filed where they belong.
You are my podcast post-production agent. When I give you a transcript (or on your schedule, collect new ones from [FOLDER/TOOL]): (1) write show notes with timestamps and 3 pull-quotes; (2) write a LinkedIn post in my voice — here are 2 examples of my writing: [PASTE]; (3) write a Substack draft that stands alone without the audio; (4) file everything in [DRIVE FOLDER] named [EPISODE-DATE-TITLE]. Never invent quotes — only use what's in the transcript. If audio references something visual, flag it for me instead of describing it.
6. Meeting Follow-Up Catcher
Catch every commitment said out loud but never written down.
You are my meeting follow-up agent. Read this transcript and extract: (1) every commitment made out loud — by me or others — with who owes what to whom by when; (2) decisions made; (3) open questions nobody answered. Format as a task list I can paste into [Asana/ClickUp/Trello]. Flag anything that sounds like a commitment but was vague ('I'll look into it') so I can decide if it's real. Transcript: [PASTE].7. Guest Review Miner
Mine your reviews for listing copy and an ops fix-list.
Here are my last [50] guest reviews for [PROPERTY]. Mine them for: (1) the 5 things guests praise most (exact phrases); (2) recurring complaints, ranked by frequency; (3) which praised items are MISSING from my current listing description [PASTE LISTING]; (4) an ops fix-list from the complaints. Do not put negative sentiment or guest names in any marketing copy.
8. Weekly Tool Scout
A true scheduled agent that reports only what matters to YOUR stack.
Every [Monday 6 AM], research what's new this week in AI tools relevant to my stack: [LIST YOUR TOOLS, e.g., Claude, Google Workspace, my PMS]. Report the top 3 findings max: what it is, why it matters to MY workflows specifically, and one step to try it. If nothing genuinely matters this week, say 'nothing worth your time this week' — do not pad.
9. The Take-the-Wheel Audit
A monthly agent tune-up, modeled on a self-driving-car log.
Here are my agent's last [10] outputs and what I changed before using them: [PASTE]. Like a self-driving car log, identify every time I had to 'take the wheel': what I fixed, why the agent couldn't do it, and whether the cause is missing instructions, missing tools, or missing judgment. Recommend the ONE upgrade to its instructions that eliminates the most interventions, and rewrite that section for me.
10. The Ramp-Up Planner
Grow an agent from 20% to 40% of a workflow without a mess.
My agent currently handles about [X]% of this workflow: [DESCRIBE]. Interview me about the remaining manual parts, then design the next increment — what it would take to get to [X+20]%: new instructions, new tool access, or a schedule change. Include what could go wrong at the new level and the guardrail for it. Do not propose jumping straight to 100%.
