How to build a blog with AI

As Generative AI becomes ubiquitous, it raises an interesting question: should Prompts used to generate content be included as part of the final piece? Not only would this serve as a form of attribution, but it would also leave a descriptive, non-deterministic mark. For this blog, it all started with a prompt from Cursor Composer using Claude Sonnet 3.5¹

Build a Next.js blog with these features:

1. Uses Markdown or MDX for content.
2. SEO-friendly setup (meta tags, JSON-LD, sitemap, RSS).
3. Scientific publication look: clean typography and footnotes.
4. Tailwind CSS (with @tailwindcss/typography) for styling.
5. TypeScript (if you think it’s beneficial).
6. Deployed on Vercel with a lightweight build process.
7. Provide instructions in a README for: Adding new Markdown posts, Editing SEO config (title, meta tags, structured data), Running & deploying on Vercel

Generate a starter project or template with clear file structure, and demonstrate a sample post to showcase the scientific style. Make sure everything is well commented so it’s easy to maintain.

I had to adjust the code to achieve the desired look, as just yet, LLMs struggle with aesthetics and/or when replicating existing designs in code. Additionally, I removed ~30% of lines of code and simplified some functionality. LLMs generally tend to overcomplicate things.

Also, when editing the content of this post, cursor autocompleted the footnote text. It was a moment of enlightenment. It's astonishing what can be accomplished with AI in 30 minutes.

It's definitely time to build.