The product
Everything between
“what should I build?”
and the first commit.
Hunting Apps finds needs people already have, researches the ones you pick, tells you whether to build, and sizes what you would build. Then it hands a brief to the tool that writes the code.
01 · Sparks
Start the day with needs, not a blank page.
Every morning the app reads places where people say what they are missing: Reddit threads, freelance postings, low reviews of top-grossing iPhone apps, Stack Exchange recommendation questions and Hacker News. Each need becomes a card with a short quote and the link to where it was said.
For each need it writes ideas, called sparks, and grades every one from 0 to 10 on competition, distribution and market size, with the searches behind the grade.
- Free to read. The feed is shared by every account and costs no credits to browse.
- 40 credits to turn a spark into a researched candidate.
- No usernames. A need is a paraphrase, a short quote and a link.
“We still take cake pre-orders by phone and lose half of them when nobody picks up.”
- Competition
- 7
- Distribution
- 6
- Market size
- 5
01b · Collisions
When strangers ask for the same thing.
A collision is several people's needs, found in different places, that one product could serve. It is the strongest kind of signal the feed has: the same problem, said independently, by people who never met.
Open one and you see each person, what they asked for, where and how often it came up. Hunt this collision starts a hunt with all of them as evidence.
- A dog groomerloses bookings when the phone rings mid-groomReddit
- A piano teacherchases parents for next term by text, one by oneA freelance posting
- A mobile barberwants clients to rebook without callingApp Store review
02 · Hunts
Seven ways to start a hunt.
A hunt is a research run. The agent writes the searches, reads the web, the app stores, issue trackers and an index of open source code, groups what it finds into markets, and checks who already sells a fix. Pick the kind that matches where you are.
Find something to build
You can build, and you have not chosen what.
Check what you have
You have an idea, a market in mind or a product already running.
Solve one part
You are building, and one piece has probably been solved already.
Typical credits for a first run. Every kind, step by step
03 · The verdict
A clear answer, with the receipts.
Every candidate ends in one verdict: worth building, worth a closer look or not worth building. It rests on three parts, each graded 0 to 10: the problem (who asked, and where), the idea (what would be built, for whom) and whether you can win (who sells it, at what price, and what their users complain about).
The rules are strict. A summary may cite only the sources it was given and name only products the market check found. Fewer than two sources, or nobody paying, holds the verdict at a closer look.
Read a recorded verdict in fullOffline Drug Deck
A drug formulary held on the device for prescribers on wards and in clinics, bought once, with no subscription and no connection.
- 13
- products found
- 8
- of them free
- 1
- buyer quoted
- 0/4
- ways to win strong
04 · Shape it
Four sizes, from one shaping.
A verdict says whether. Shape it says how big. From the people and products on file, it lays out what you would build at four sizes: a niche that wins on one thing, a first version, a size that grows towards the market, and one level with the leaders.
Each size is graded on reach, edge and effort, and the one with the best net value that is not a bet is recommended. Tick your own mix and it is measured the same way.
A real one: the Honey Jar Kit, v1 recommended- Reach
- Edge
- Effort
- Reach
- Edge
- Effort
- Reach
- Edge
- Effort
- Reach
- Edge
- Effort
05 · Project and hand-off
Keep the reasoning with the build.
Take a size to a project and it becomes a workspace: the plan, the research, your own notes and files, and a chat that knows all of it. Each project carries a thesis and a stop condition, so you know what would make you walk away.
When you are ready, the hand-off drafts a build brief and pushes it to a new repository in your GitHub account, with guide files your coding tool reads first. Or connect Claude, Cursor or another tool over MCP and let it read the project directly.
- Claude Code
- Cursor
- Codex
- Replit
- Draft the briefFrom the plan, the research and your files
- Connect GitHubYour own account
- Create the repositoryPublic, and the page says so first
- Build itREADME, AGENTS.md, CLAUDE.md, the brief and a build plan
Built not to make things up
Every claim can be checked.
Research you cannot check is research you have to trust. These rules are in the code, not in a promise.
Each page the agent reads is saved to your library with the search that found it.
A link in a result must be one a source returned. Nothing is made up along the way.
Page text is fenced as untrusted, and a turn that read it cannot follow a link it suggested.
The one step allowed to invent badges its ideas, and they never count as evidence.
Read tomorrow's sparks,
or hunt your own idea today.
200 free credits. No card required.