My results for an AI building challenge on almost no budget

I've been building projects I find interesting with AI tools since the beginning of the year, and lately I've been focusing on increasingly complex software, partially just as a challenge to myself, and partially to determine at what point the AI tools start to fail in the area of software development.

For a recent project, I picked an intentionally too-ambitious goal: can I recreate most of a complex business application (in this case, Excel) in a week, for less than $100? How far can I actually get on that?

After planning it out pretty thoroughly with Claude, and brainstorming some additional AI enhanced features (e.g. plain english explanations for complex formulas and error diagnostics), I started on the implementation using the most recent Opus model. The bulk of that work only took a few days, with a few additional days of testing and polishing.

I'm pretty happy with the results (the name HolySheet occurred to me and seemed pretty apt, so I went with that)... you can check it out for yourself here: https://www.holysheet.dev

Features Built

I was able to build out most of the spreadsheet basics you'd expect:

• File manipulation
• Formulas
• Rich cell formatting and styling
• Pivot tables and charts
• Multiple sheets

I don't want to just do a feature dump here in this post, but the app is pretty full featured.

Thoughts on AI Costs

Anyway, it does make me wonder where all the token spend I hear about in the tech news is actually going... like if one person can create a useful, complex business app in a few days for something like $50, what exactly are companies spending $500M on? Honestly it's a bit baffling. (I realize this is simplifying things, but I think my point is still valid.)

Closing

At any rate, I welcome any feedback and comments on this project. And if you'd like to collaborate on an interesting software project, please do reach out.

-Jeff