Interestingly, he prompts for "an app", and the agent proposes "using Flask and Vanilla JS".
This is a somewhat strange implementation choice, as there's no need to create a `/resize` API endpoint - this could be implemented entirely in JS as a static site (enabling it to be hosted by free on somewhere like GitHub Pages, avoiding the issue that "I deployed it here . . . But it's costing me a bit of money so I don't expect it'll be around for too long.").
The implementation also has bugs like trying to validate an SVG file by checking that it starts with `<svg`, when SVG files often begin with something like:
Thousands of these kinds of tools would fit in a few megabytes. Seems like we could easily make a GitHub Pages site with every random word counter anyone could ever want, using Web Workers so it's all available offline.
You could probably even have an in-browser LLM frontend for it for less structured tasks.
You could even have an integrated editor for making new tools with AI and creating a pull request to share them.
Interestingly, he prompts for "an app", and the agent proposes "using Flask and Vanilla JS".
This is a somewhat strange implementation choice, as there's no need to create a `/resize` API endpoint - this could be implemented entirely in JS as a static site (enabling it to be hosted by free on somewhere like GitHub Pages, avoiding the issue that "I deployed it here . . . But it's costing me a bit of money so I don't expect it'll be around for too long.").
The implementation also has bugs like trying to validate an SVG file by checking that it starts with `<svg`, when SVG files often begin with something like:
<?xml version="1.0" encoding="UTF-8" standalone="no"?> <!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
Thousands of these kinds of tools would fit in a few megabytes. Seems like we could easily make a GitHub Pages site with every random word counter anyone could ever want, using Web Workers so it's all available offline.
You could probably even have an in-browser LLM frontend for it for less structured tasks.
You could even have an integrated editor for making new tools with AI and creating a pull request to share them.