Hey, guys!
Let me show you how I build new hobby websites with AI tools nowadays.
🧠Brainstorm with AI agent
Using a voice transcription AI tool, like Cursor's native recorder or "Wispr Flow", I brain dump all my ideas and requirements into text and save it in a PRD.md file or in a Google Doc. I also attach a bunch of other copy-pasted text that might be related to that project.
🧱 Init a codebase
I personally go with npx create-next-app@latest my-next-app --tailwind --typescript --eslint --app --turbopack and
--use-pnpmnpx shadcn@latest init.
Then, I use an AI Agent tool, such as Claude Code, to create the project. I paste the document create before into a prompt and add some extra details, for example:
Project in nextjs, tailwind, shadcn components (don't create them, import them using shadcn commands), @neondatabase/serverless Neon DB, @stackframe/stack Stack Auth
Full Prompt: https://gist.github.com/Markkop/765612eecb7f6a7e1c2564eda03b228b
⚒️ Fixes and features
Now, using Claude Code and Cursor, I finish the integrations I want and the fixes needed. I also try to leverage AI at max via Playwright MCP/Chrome MCP/Cursor (browser testing) and Neon MCP (db updates).
This is where I most leverage my dev knowledge: I guide the AI to build the features I plan, change libraries, follow patterns and so on.
🌎 Publish and Release
Finally, when I have something okayish to release, I setup Vercel and domains.
When I started writing this post, I realized there's a lot of more profound knowledge applied to the process that is too much for write in a single post, so take this more as an overview of the process
Top comments (0)