Someone asking
real demand- A one-star review.
- A GitHub issue people keep upvoting.
- A thread that starts with “I’d pay for something that…”
The problem
Building got cheap. Choosing didn’t.
You can ship a working product in a weekend now. So can everyone else.
That moved the hard part. The question used to be “can I build this?” Now it’s “will anyone pay for it?”, and most of us still answer it the old way. We go with whatever sounds good, or we ask an AI.
Then we spend three weekends building it and launch to silence.
The test
Ask three AIs for product ideas. Count how many are different.
Give me 10 product ideas a solo developer could build in two weeks and sell online.
Claude
- Notion template pack
- Invoice generator for freelancers
- AI resume builder
- Productivity Chrome extension
- SaaS starter kit
- Habit tracker app
- Link-in-bio page builder
- Pomodoro focus timer
- Screenshot beautifier
- Simple uptime monitor
ChatGPT
- AI resume builder
- Notion template pack
- Habit tracker app
- AI social caption writer
- SaaS starter kit
- Invoice generator for freelancers
- Productivity Chrome extension
- Pomodoro focus timer
- AI meal planner
- Expense tracker app
Gemini
- AI resume builder
- Habit tracker app
- Notion template pack
- AI social caption writer
- Productivity Chrome extension
- Link-in-bio page builder
- Invoice generator for freelancers
- Printable digital planner
- AI YouTube title generator
- QR menus for restaurants
Sample answersAll threeTwo of threeJust one
That is what these models are built to do: give the most likely answer. The most likely answer is also the one every other builder is getting. If your idea came out of the same prompt as everyone else’s, you are starting in a crowd.
You can run this yourself in five minutes.
The idea behind it
Start from receipts.
A good idea leaves evidence before anyone builds it. Better idea looks for two kinds, and it wants both.
Someone paying
real sales- A product that already sells in that space.
- Even a clumsy one.
- Especially a clumsy one.
Both? It passes.
When both exist, you are no longer guessing whether the problem is real. You are deciding whether you can do it better for a specific group of people.
No receipts, no idea.
If the skill can’t point to evidence, it doesn’t show you the candidate.
How it works
What happens when you run it
- 01
It asks about you.
What you can build, how much time you have, and who you can actually reach. An idea that needs an audience you don’t have isn’t your idea.
- 02
It searches cases, not its imagination.
Candidates come from collected cases: small products that sold, launches that flopped, and complaints nobody has fixed.
- 03
It argues against every candidate.
No sign anyone pays. Too crowded. Wrong buyer for you. Most candidates don’t survive this step, and you see why each one was cut.
- 04
It ranks what’s left with plain scripts.
The scoring is code you can open and read, not a hidden prompt.
- 05
You get a short list with links.
Each idea is one card: what to build, who’s asking, who’s paying, how it could fail, and the sources with the date they were checked.
What you get
What’s in the box
Real cases
Each with a source link and the date it was checked.
Wins and failures
Products that sold, products that sold a little, and products that got applause and no sales. The flops are there on purpose.
Demand signals
- One-star reviews
- GitHub issues
- “I’d pay for…” threads
- Posts from people stuck trying to ship something
The kinds of products solo builders actually sell
Small paid utilities and CLIsAdd-ons for creative toolsExtensions for developer and note-taking toolsProducts for the AI-coding ecosystemOne-feature appsWeekend buildsData products
A kill review
It argues against each idea before you see it.
Scoring you can read
Search, selection and ranking run as scripts in the skill folder.
An offline viewer
One HTML file with every case in it. Open it in a browser. No account, no connection.
Three tools, one install
Claude Code, Codex and opencode.
Nothing to host
It runs inside the agent you already use.
- No server
- No API key
- No account with me
Who it’s for
Made for people who can build but haven’t picked what
It fits if you
- can ship a small product in a week or two
- already use Claude Code, Codex or opencode
- would rather sell something small that people want than pitch something big that might work
It doesn’t fit if you
- want a venture-scale startup idea
- want a guaranteed outcome
- have never used an AI coding agent
Compared with what you do now
Four ways to pick an idea, and what each one costs you
01
Ask a chatbot.
Fast and free. Also the same list everyone else got.
02
Follow trend lists.
They show what’s loud this week, not what people pay for.
03
Subscribe to idea newsletters and databases.
Someone did real research, then sent the same ideas to every subscriber. None of it knows what you can build or who you can reach.
04
Do the research by hand.
Reading reviews, issues and threads works. It takes hours per idea, which is why most of us skip it.
Better idea is the by-hand method, done for you, inside the tool you already build in, starting from your own constraints.
What it won’t do
Worth knowing before you buy.
It won’t promise revenue.
Nobody can. It shows evidence that a problem is real and that people pay in that area. Whether your version sells is still up to you.
It won’t replace talking to buyers.
It gets you to a short list worth testing. You still test.
It isn’t live market data.
Every case shows the date it was checked. Markets move. Open the links.
It isn’t for big startups.
The data is about small products one person can ship.
Pricing
One price. Everything in it.
Better idea
$39one time
- All the cases
- The offline viewer
- Installs from one pasted line, in all three tools
One payment. No subscription. Secure checkout by Polar, which handles tax and emails your receipt.
If it stops you from spending one weekend on the wrong thing, it has paid for itself.
Not for you? Email me within 7 days for a full refund.
Questions
Which tools does it work with?
Claude Code, Codex and opencode. It’s a standard SKILL.md skill, so the same folder works in all three. Your agent installs it in the right place for its own tool.
Do I need an API key or an account?
No. The skill runs inside your agent and uses whatever plan you already have. There is no server on my side.
What do I need installed?
Python 3.10 or later. The helper scripts use only the standard library, so there is nothing else to install.
How is this different from just asking Claude for ideas?
Without the skill, the model answers from what it absorbed in training, which is why everyone gets similar lists. With the skill, it has to start from specific cases with sources, score them against your situation, and throw out anything it can’t back up.
Won’t everyone who buys this get the same ideas?
The short list starts from your answers: your stack, your time, your audience. Two people with different answers get different lists. Two people with identical answers will see overlap, which is why every idea shows its sources and its weak points instead of a verdict. You decide, with the evidence in front of you.
Where does the data come from?
Public sources: product pages, makers’ own sales posts and post-mortems, reviews, GitHub issues and community threads. Cases are stored as facts, short excerpts and links, not copies of the originals.
How current is it?
Every case carries the date it was last checked.
Does my data go anywhere?
Not to me. There is no server and no tracking in the skill. Your prompts go to your AI provider the same way they always do.
What if I build one of these and it doesn’t sell?
That can happen. The skill lowers the odds of building something nobody wants. It doesn’t remove them.
Can I get a refund?
Yes. Within 7 days of purchase, no questions asked: email me and I’ll refund the full price. The refund policy has the details.
Do I get an invoice?
Yes. Polar is the merchant of record. It emails a receipt and handles VAT or sales tax for your country.
How do I install it?
Open Claude Code, Codex or opencode and paste this. Your agent unzips the download and runs the installer for its own tool, which installs the skill once for all your projects and checks every copied file. It usually takes about a minute. Restart Codex or opencode afterwards; Claude Code usually picks it up right away.
Unzip the newest better-idea ZIP in my Downloads folder into a new temporary folder, then follow better-idea/INSTALL.md inside it to install the better-idea skill globally, for all my projects.
Why I built this
The code was never the problem. I could build every one of them. I just kept choosing what to build the same way: I picked what sounded good to me.
So I started collecting the boring evidence instead. Small products that actually sold. Launches that got applause and no sales. Complaints people keep posting that nobody fixes. The pattern was hard to ignore: what sells for solo builders is rarely clever. It does one job, for people who were already paying for a worse version.
Better idea is that collection, turned into something my coding agent can use before I write a line of code.
Pick the next one with evidence.
Run it once before your next build. A few minutes with the short list costs less than three weekends on a guess.
See what’s included