1. Blog
  2. Shipping

Shipping solo with AI: how my games go from a folder to a live URL

How I, one software engineer, use AI to take a finished project from a local folder to a live website in a couple of minutes, and what you can copy.

Shipping solo with AI means one person doing the whole job, from idea to live website, with an AI assistant handling the repetitive steps. For me that job is: write the thing, put it on GitHub, deploy it, give it a web address, and list it where people can find it. I now spend about two minutes on that last stretch for each new game. Here’s how it works and what you can borrow, even if you’ve never deployed anything.

The problem: building is fun, shipping is chores

When I build a browser game like Keycraft Arcade or Kaiju City Smash, the fun part is the game. The part that used to eat my evenings came after:

  • create a GitHub repository and push the code
  • create a hosting project and connect it to the repo
  • pick a web address, add a DNS record, wait for HTTPS
  • make a cover image, write a description
  • add the game to my arcade site so people can find it

None of this is hard. It’s just ten small steps, and missing one means the site is broken or invisible. That’s exactly the kind of work AI is good at, as long as you write the steps down once.

The idea: write the recipe once, let AI run it

Instead of asking the AI to “deploy my game” from scratch each time, I wrote the recipe as a skill: a Markdown file that explains the steps, plus a script that does them. Now I say “ship Kaiju City Smash” and the assistant:

  1. Proposes a short name (kaiju), a one-line description and tags. I say “ok”.
  2. Takes a screenshot of the running game for the cover image.
  3. Runs a check first that changes nothing, so problems show up before anything is created.
  4. Creates a private GitHub repo and commits the code.
  5. Creates a Vercel project linked to that repo, so every future git push redeploys automatically.
  6. Gives the game its own address, like kaiju.play.karankharode.com, and waits until HTTPS works.
  7. Adds the game to Play Arcade, my portal, which rebuilds itself whenever a game changes.
  8. Opens both pages in a browser and checks them before telling me it’s done.

My part: say the name, approve a few permission prompts, and sometimes click one setting on GitHub that the AI isn’t allowed to touch.

Why this works (and what usually goes wrong)

Plain-words instructions beat clever prompts. The skill file reads like notes for a new teammate: what to do, in what order, and what never to do (never make a repo public, never use a different commit author, DNS records must be “DNS only”). The AI follows rules better when they’re written down than when I remember to say them.

A dry run before the real run. The --check step catches the boring failures (logged out of a tool, a missing cover, a build error) before anything half-exists.

One-time setup, done once. A wildcard DNS record means every new game gets an address without me touching DNS again. Connecting Vercel to GitHub once means no clicking around per project.

Small, reversible steps. Everything lives in Git, so if a deploy breaks, I promote the previous version in one click. That safety net is why I’m comfortable letting AI do the work.

What you can copy today

You don’t need my exact setup. The pattern is what matters:

  • Do it by hand once with your AI assistant, and ask it to write down every step it took.
  • Turn those notes into a Markdown file in your project (a README, a CLAUDE.md, or a skill). Include the rules you’d tell a new teammate.
  • Put your project on GitHub. It’s the undo button for everything AI does, and every hosting service plugs into it.
  • Use a host that deploys on push, like Vercel or Netlify, so “shipping” becomes “saving”.
  • Ask the AI to check its own work in a browser before it tells you it’s finished.

What’s next

I started shipping with games because they’re fun to show, but this is how I build everything now: apps, sites, this blog. I’ve since written about what vibe coding actually is, how to do it better, and why I think everyone, not just programmers, should learn GitHub. The pipeline itself has a write-up on the Play Arcade project page.

If you want to see the result of this pipeline, go play something. It took me longer to write this post than to ship the last game.

shipping solo with AIvibe codingdeploymentGitHubVercel

Keep reading

All posts →

One JSON file in, one finished reel out: how I render short videos with code

How I built a local pipeline that turns a JSON script into a finished vertical video: local text-to-speech, word-level timing, a web page as the renderer.

Shipping4 min read

What is vibe coding? A plain-words guide, and where it goes wrong

Vibe coding means building software by describing what you want to an AI and letting it write the code. Here's how it works, who it's for, and its limits.

Vibe coding4 min read

How to vibe code better: 12 habits that stop AI from breaking your project

Twelve practical habits for building software with AI: small steps, commits before every prompt, reading the file list, written rules and real checks.

Vibe coding4 min read