For $20 a month, you can build sites and apps yourself.
The thirty-plus works on this site — study sites, an Android app, browser games, a lottery that runs on a real quantum computer — were all made by talking to Claude Code, a "coding agent", in plain language.
I am not a programmer; I studied humanities. This page says honestly what it can do and what it cannot, with real examples.
The line, drawn up front. Claude Code is not magic. It makes mistakes, it overreaches when left alone, and you still have to check what it made.
Even so, the boundary between "people who can build" and "people who can't" has genuinely moved.
What a coding agent is
A chat AI like ChatGPT stops at showing you code. A coding agent goes on from there. Inside your own computer it creates files, runs commands, reads the errors and fixes them, runs the tests, and publishes.
Your part is to say what you want, in your own language, and to answer the questions it asks along the way.
You: "Make a memory-matching app for learning people's names. I want it on Android." Claude Code: creates a Flutter project, builds the screens, grounds the drill rules in memory research, builds it, and produces the package for the Play Store.
That is not hypothetical; it is how Namaegao came to exist.
What it costs
Claude Pro
$20 a month. Includes Claude Code. Usage is capped and you may wait at busy hours. Start here
Claude Max
$100 or $200 a month. Much larger caps, for when you are building for hours every day
Hosting
Vercel and Cloudflare both have free tiers that cover a personal site. This whole site runs on Cloudflare's free tier
Your own domain
About $10–15 a year. Optional — you get an address like xxx.vercel.app for free
Prices change. Check the official pricing page before you sign up. This page reflects September 2026.
What it can do (examples from this site)
Websites — this page, the Genius Ranking, the Multiverse Guide. Say "quantum mechanics for humanities readers in fifteen chapters" and it builds the outline, the text, the figures and the search-engine plumbing
Phone apps — Namaegao (on Google Play), Drowsiness Guard (watches your eyelids through the camera). You never write Flutter or Kotlin yourself
Games — Robot vs Drone vs Human plays with your phone's accelerometer and microphone; Furo Arena is for the bath. All in the browser
A real quantum computer — the Quantum Draw picks winners from bits produced by an IBM quantum computer. Figuring out the API was left to the agent
Things that keep running — BCI WATCH and the paper ranking collect papers and news every morning. "Collect at 7 every morning and email me" is enough
Publishing — it runs the Vercel or Cloudflare commands itself, checks the live URL, and only then says "done". Committing to GitHub can be delegated too
What about images, sound and video?
This is where most misunderstandings live, so plainly: Claude Code is not an image or video generator.
What it can do is write and run programs that make images, sound and video. That goes surprisingly far.
The six alarm tones in Okiro! Alarm were synthesized by a Python script it wrote — no stock sounds
Yes video processing
Pagedive cuts a video of someone flipping a book into one PDF page per leaf. With ffmpeg or OpenCV code it can cut, join, burn subtitles, make timelapses
Not directly photo-like images, acted video
There is no image model inside Claude Code. Either have it write code that calls an image-generation API (paid separately), or hand it assets made elsewhere. Ask it to "draw" and you get what code can draw: shapes, patterns, simple illustrations
2 Install the Claude desktop app and open the "Code" tab. If the terminal does not scare you, this works too:
npm install -g @anthropic-ai/claude-code
3 Make one folder for the project and type claude there (in the app, just pick the folder)
4 Ask, in your own words. Start small:
"Make a one-page personal site: my name, what I do, how to contact me. It should look good on a phone."
5 When it works: "Publish it on Vercel." The first time you will be asked to create a Vercel account and log in yourself. A few minutes later you get a URL
What makes it go well
Ask small, then add. "The whole app at once" gets lost halfway. One page, then two, then a feature
Say "verify it". Don't stop at "built"; ask it to open the live URL and check, or write tests and pass them. Agents verify when told to
Write your rules in a file. A file called CLAUDE.md with "never publish my real name", "never delete without asking" is read before every session. This site's policies live there
Keep secrets away from it. Passwords and API keys you enter yourself. A well-built agent stops and says "you type this one"
Read what it made. Mistakes happen, especially in facts — numbers, dates, sources. Open the links yourself
What it cannot do, or should not
Honestly: app-store accounts and review are human work (identity, addresses). Payments you make yourself.
A production service for many people (billing, personal data) deserves one look from someone experienced.
And deciding what to build is still the hardest part, and it will not do that for you.
More on how this site is made
AGENT NOTES — field notes from people using Claude Code, collected daily
Agent Optimizer — how to write instructions, the shape of a good CLAUDE.md