Three Steps to Your First Output with AI — What to Decide Before Picking a Tool

You have already signed up for the AI tools. Yet you open the window and just stare at it for a long time.

It's because you don't know what to ask it to do.

I've had more than 500 one-on-one conversations with aspiring founders, and the question I hear most is "Which AI tools do you use?" But the real sticking point was never the tool. It was that they hadn't decided what to make first.

Switching tools won't fix that. The order is backwards.

Why people stall at the start

There's a scene I see over and over in the field: someone asks the AI, "Write a plan for our service."

An output comes back. It looks plausible. But there's nowhere to take it. It's not going to an investor, it's not being handed to a developer, and for your own reference it only says what you already know.

So it gets saved and never opened again.

The same pattern shows up when I review students' work in my classes. The screens are drawn, but there's no answer to questions like "When does this screen appear?" or "What happens if it fails?" Something got made, but it isn't a document that lets someone decide something.

They share one thing: it was made without deciding who would receive it.

Step 1. Name the one thing you'll have in hand this week

"Using AI" is not a goal. The goal has to be the name of an output.

• Don't write it like this — Try preparing my business with AI
• Write it like this — This week, make "a one-page requirements sheet for the meeting with the outsourced dev agency"

Once the output has a name, what to tell the AI follows naturally. Without a name, all you can come up with is a vague instruction like "write a plan," and vague instructions get vague answers.

When I was planning a service in the CrossFit space, I set the goal as "one page of questions to ask 20 gym owners." I actually met those 20 people, and based on those interviews I went on to build an MVP delivered to 10 gyms.

Rather than researching the market broadly, deciding on one thing to hold in hand was faster.

To do today — Rewrite what you're working on in the form "one page of ___." If it doesn't fit that form, it's still a topic, not a task.

Step 2. Write down who will receive that output

Without this, the format never settles. Even the same "requirements" become a completely different document depending on who receives them.

• Outsourced dev agency — needs screens, states, exceptions, and data sources. Without them, you either get no quote, or the quote changes later.
• Review panel — needs the problem, the customer, evidence, and numbers. Without them, you get stuck at "So why would this work?"
• Customers (for validation) — need interview questions and behaviors to observe. Without them, you only confirm what you wanted to hear.
• Yourself (for organizing) — need a clear split between what's decided and what isn't. Without it, you start thinking from scratch every time you reopen it.

The moment you write down the recipient, your instruction to the AI becomes specific. Instead of "write the requirements," it becomes "write this so an outsourced dev agency can give a quote, including the states of each screen and what happens on failure."

When I ask "What do you want to make?" in one-on-one consulting, most answers are something to hand to the next person: a business plan, materials for discussing outsourcing, a pitch deck. An output is less an endpoint and more a handoff asset.

To do today — Next to the output you named in Step 1, write down one recipient. If there are two or more, it's better to split the document.

Step 3. Finish a draft within 30 minutes, and mark where a person needs to fix it

If you try to finish it, you can't start. The goal is not completion but a state that can be reviewed.

The method is simple. Get a draft from the AI, then read it yourself and mark three things.

1. Confirmed — facts you can back up with evidence
2. Needs checking — plausible but without a source (especially market size and competitor figures)
3. I must decide — things the AI must not decide for you (price, target, features to give up)

Number 3 is the core. AI writes sentences well, but it can't decide what to give up. That belongs to the person running the business.

This is especially true when you ask it for market research. Numbers come out, but often without a source or reference date. Put those numbers straight into a business plan, and they collapse at the review table under a single question: "Where does this figure come from?" I've seen people get stuck at exactly this point many times in consulting.

It's better to remove numbers that have no source. A blank is safer than a wrong number.

To do today — Use three highlighter colors on your draft. If much of it is uncolored, you haven't actually read it yet.

One sentence that runs through all three steps

Getting to your first output with AI is not a matter of picking the right tool.

It's a matter of deciding first what to make, for whom, and how far you'll make the calls yourself.

• Step 1: the name of the output — without it, your instructions get vague
• Step 2: the person who will receive it — without it, the format never settles
• Step 3: the points you decide — without them, all you're left with is a plausible-looking document

Once these three are set, you'll get results whatever tool you use. Without them, no matter how many times you switch tools, you'll end up staring at an open window.

If you do just one thing today

Write three lines on a sheet of paper.

1. What I'll make this week: one page of ____________
2. Who will receive it: ____________
3. What I'll decide myself: ____________

If all three lines are filled in, you can start today. If even one is blank, that blank line is exactly where you're stuck right now.

---

Not sure where to start?
A 2-minute diagnostic lays out the one task to tackle first, a plan for starting a 2-week pilot, and the risks a person must check. Answer only as much as you're comfortable with.

STAR-T AI Business Operations Diagnostic →

---

Notes

① References and verification date
This article does not cite external statistics or research. It is based on 500+ (cumulative) one-on-one consultations with aspiring founders and patterns repeatedly observed in classes. The CrossFit case in the text is a service planning project carried out in 2023; the interviews with 20 gym owners and the MVP delivered to 10 gyms are the actual figures from that time. Verified 2026-08-04.

② Figures not used
Figures from internal surveys and CS responses were deliberately left out. The samples are small, and expressing them as percentages could be misread as market statistics. There are also no direct quotes from individual consultations or surveys.

③ How this was made
The draft was produced with AI, then verified and edited by a person. Before publication it went through three reviews: facts, tone, and legal.