Build the Small Tool You Keep Wishing Existed

I'd ignore every post telling you to build an AI side hustle. Build the small tool you'd use yourself, and decide later, once you have something that works, whether anyone else wants it.

You've already got an idea. Maybe it's splitting the bills with your housemates or working out which shift pattern pays best.

Or it's turning the kids' school emails into a calendar, or changing a recipe for six into one for two.

Each time, you've thought "there should be an app for this", and then you carried on doing the job in your head.

Now there can be an app for it, and you can make one in an afternoon without writing code.

Claude and ChatGPT can both build a working web page from a plain description, and they can put the page at a real link you can open.

Catch the idea in your own week

Don't ask the AI for app ideas, because you'll get a to-do list app.

Instead, keep notes on your phone for seven days. Write something down every time you do a sum on your fingers or open the same spreadsheet to look up one number. Do the same when you mutter "there should be a thing for this". Then paste the week's notes in.

Here are my notes from one week. Each note is a time I worked something out by hand, looked something up again, or thought "there should be a thing for this". For each one, say how often it happens. Say how simple it would be as a one-page web tool. And say whether it needs any information I don't have. Then pick the one that's most annoying and simplest, and tell me why. Don't build anything yet.

The good ideas are embarrassingly small. "I recalculate the dog's medication dose every time the vet changes it" makes a better tool than anything you'd call a business.

If the week turned up nothing, or everything on the list is too big, then let AI interview you. Try this. "Ask me about the small repetitive jobs in my week, one question at a time. I mean the things I work out by hand or keep in my head. Stop when you've found three."

People rarely notice their own Tuesday jobs until someone asks.

Pick the job you'd use the tool for on Tuesday, which is rarely the biggest one.

Describe the smallest version

Write down what the tool does in three sentences a friend would understand, and hand those sentences over before anything gets built.

I want to build this. [your three sentences]. Before you build it, write back the simplest possible version. That means what I type in, what it shows me, and nothing else. List what you're leaving out and why. Then ask me three questions about how I'd actually use it.

Answer the AI's questions and cross out anything you won't need on day one.

Logins, accounts, saving information online and payments can all come later if you ever need them, and for most personal tools you never will.

Could you explain the tool to someone while the kettle boils? If not, the tool is still too big.

Keep those three sentences somewhere safe, because they're the real source of the tool.

Have it built and published

In Claude, ask for the tool as an artifact, which is Claude's word for a page it makes for you, and then ask Claude to publish it. In ChatGPT, ask for the tool as a site.

Either way, you get a page at a real link. The page stays private until you share it, so nobody sees the rough version but you.

Build the simplest version we agreed, as a single web page I can open on my phone. Keep the design plain. When it's done, tell me the three things most likely to be wrong with it so I know what to test first.

Test it like you don't trust it

The AI builds whatever you describe, including any mistakes in your description, so put in a number you already know the answer to and check the result by hand.

Then try something silly, such as a negative number, an empty box or a word where a number should go, and watch what happens.

When something's wrong, say so in plain words. For example, "when I put in 6 people it shows the total for 5." You don't need to know why it went wrong, because the AI will fix the tool and you test again.

Use it for a month before you show anyone

Add the tool to your phone's home screen and use it every time the job comes up.

Note what annoys you, such as a box that's too small or an answer you still double-check.

By the end of the month, one of three things will be true. The tool might be useful and finished, and you've quietly got an hour a week back. It might be useful but missing one thing, which you now understand well enough to ask for properly.

Or you've stopped using the tool, which is also an answer, and a cheap one.

Pages the AI builds can quietly break when Claude or ChatGPT gets updated. But your three sentences are still where you saved them, and you can rebuild the tool from them in ten minutes.

Leave money until last

Wait until a friend sees the tool and says "can I have that?" Only then is it worth asking whether anyone else would want it.

A tool that works for you has to work for strangers too, on their phones and with their information, and they'll email you when it doesn't.

Taking payments means a payment company, terms and tax. None of that is impossible, but it's a different afternoon from this one.

Building is now the easy part. Finding people who'll pay is the same hard work it's always been, and no prompt changes that.

Anyone promising "£200 a day on autopilot" is selling you the prompt, not the result.

Keep the first tool to maths and words, too. Tools that store people's details, take money, or handle health or legal information need more care than an afternoon.

AI builds the page. You know which Tuesday job deserves one.