Four jobs before lunch
A normal Tuesday for me: 9am I'm a developer, 11am I'm writing landing page copy, 1pm I'm on a discovery call pretending I have a sales process, 3pm an angry email lands and I'm support, 5pm I'm doing the invoicing I've been avoiding since Friday.
The work isn't the expensive part. The switching is. Every hop lands me in front of a blank page in a role I'm mediocre at, and I lose fifteen minutes just working out what good even looks like here. Four hops a day, five days a week. That's most of a working day a week spent staring.
So I did what everyone does now: I asked an AI assistant. "Help me write a cold email." "How should I price this?" And I got exactly what you'd expect — competent, structureless mush. Advice that sounds like a LinkedIn post. The problem wasn't the model. It was that I was asking like someone who didn't know the domain, because in that particular hour, I didn't.
The fix turned out to be boring: stop asking questions, start pasting briefs. A prompt that names a framework, hands over real numbers, sets constraints, and demands a decision rule at the end gets back something I can act on. One that doesn't gets a fortune cookie.
I kept the ones that worked in a file. That file is now 54 prompts across the six jobs I actually do — validation, marketing, sales, ops, finance, support — and I've packaged it up. Here are three of them in full, so you can judge the pattern yourself.
1. The prompt I run before I build anything
When to use this: when you're in love with an idea and need a hostile reviewer to try to kill it before the market does.
Act as a skeptical seed-stage investor who has seen 1,000 pitches and passed on
most of them. Here is my idea:
[DESCRIBE THE IDEA IN 3–5 SENTENCES: who it's for, the problem, the solution,
and how it makes money]
Tear it apart. Specifically:
1. Name the 3 assumptions most likely to be fatally wrong, ranked by
(likelihood of being wrong) × (damage if wrong).
2. For each, describe the cheapest test (under $100 and under one week) that
would produce real evidence either way.
3. List who already solves this and why customers might prefer the status quo,
spreadsheets, or doing nothing.
4. Tell me what evidence would change your mind from "pass" to "curious."
Be direct. Do not soften the critique and do not end with encouragement.
Why it works: the ranking formula stops the model listing ten generic risks and forces a prioritisation. "Under $100 and under one week" converts criticism into this week's to-do list. And the last line matters more than it looks — without it you get a paragraph of encouragement at the end that quietly undoes the whole exercise.
2. The prompt that gets my week back
When to use this: when your week disappears into repetitive admin and you need to know what to automate first.
You are an automation consultant. I'm a solo founder. Here is how I
actually spent my time last week, with hours per task:
[PASTE YOUR TASK LIST, e.g. "answering support emails 4h, invoicing 1h,
scheduling social posts 2h, sales calls 5h, writing content 4h…"]
1. Score every task on two axes: repetitiveness (how rule-based is it)
and founder-necessity (does it need MY judgment/relationship).
2. Sort all tasks into four quadrants: automate now, delegate, keep
but time-box, and eliminate.
3. For the top 3 "automate now" candidates, rank by ROI =
(hours saved per month) ÷ (setup effort), and describe the
automation for each: trigger, steps, tools ([MY STACK, e.g. "Gmail,
Stripe, Notion, Zapier/Make"]), and failure mode to watch.
4. Estimate total hours recovered per month and tell me the ONE
automation to build this week.
Why it works: it runs on your logged hours, not a hypothetical founder's, so the output can't be generic. And step 4 is the whole point — a list of twelve automation ideas is another form of paralysis. I want one.
3. The prompt for the email I shouldn't send at 11pm
When to use this: when a furious email lands and your first-draft reply would make things worse.
You are a customer-success lead who specializes in saving angry
customers. My product: [PRODUCT DESCRIPTION]. Here is the customer's
message:
[PASTE THE ANGRY MESSAGE]
And here is what actually happened / what I can offer: [FACTS + WHAT
YOU CAN DO: refund, fix timeline, workaround, nothing].
Write my reply:
1. First, tell me privately (not for the customer): is this customer
right? What did we likely get wrong?
2. The reply itself, under 150 words: acknowledge the specific
failure without legal-speak or "we're sorry you feel that way",
take ownership of what's genuinely ours, state what I'm doing
about it with a concrete action and date, and make the remedy
(if any) without them having to ask again.
3. One sentence I must NOT send (the defensive thing I'm tempted to
say) and why.
4. A follow-up note to send [N] days later checking the fix landed —
this is where angry customers become loyal ones.
5. The product/process fix I should make so this exact ticket never
happens again.
Step 1 and step 3 are why I still use this one. The private read on whether the customer is right usually deflates my defensiveness before I've written a word, and naming the sentence I'm tempted to send is unnervingly accurate about every time.
They're also built to chain: interview notes from the Mom Test script feed the testimonial-to-copy miner, whose output feeds the landing-page hero prompt. Outputs become inputs.
Who this is not for
- If you already have a playbook, skip it. Someone who's run sales or done pricing work professionally will find their own frameworks in here with the serial numbers filed off. The pack is for people covering six roles badly, not one role well.
- It's prompts, not software. Markdown files plus one print-ready HTML page. Nothing runs, nothing integrates, no accounts. If you want a tool, this isn't one.
- The placeholders are work. Every prompt wants real numbers, real quotes, real transcripts. Fill them with vagueness and you get vagueness back — the specificity is doing the heavy lifting, not the wording.
- It won't make the decision for you. Several prompts deliberately end in a scorecard or a pass/fail rule. That's a structured argument to react to, not an answer.
Getting it
The Solo Founder's AI Prompt Pack is $19 on Gumroad: https://afeldman2.gumroad.com/l/zdyxz?utm_source=devto&utm_medium=article&utm_campaign=promptpack-launch
54 prompts across six categories, organised Markdown for daily copy-paste plus a single styled HTML file you can print to PDF. Works with any assistant that takes a long prompt — ChatGPT, Claude, Gemini, whatever you already pay for.
And if you just take the three above and never buy anything: genuinely fine. The pattern — name a framework, hand over real data, constrain the output, demand a decision rule — is the part worth stealing.
Top comments (0)