[pixel_popping]

You can have AI do way more automated testing in a loop, let cheap models run 24/7 so all edge cases can be found over time by pulling random repos, you must make sure the agent is entirely autonomous, ask another agent to make a list (also in a loop so it keeps appending to it) and another agent that try to fulfill "run it with Deno".

[jdxcode]

I think node already does this (sans AI), idk where the code lives but deno could probably just use the same list of repos

[someguyiguess]

You're getting downvoted because "AI bad" but this is actually a good idea.

[jorvi]

Because none of that requires "AI". You can use a script to clone GitHub top NPM repos (by whatever measure), and then 'simply' try to migrate them to Deno and log all the errors and warnings. Maybe you compile those errors in a neat list with an LLM, and ask it to emphasize things it considers important. But even that can be done inside the script with loglevels and maybe some regex.

Parent is getting downvoted because they're suggesting something that is akin to "why are you running this simple, reliable, efficient Python script instead of feeding inputs to an LLM and praying you get the outputs you want".

Respect yourself by respecting your craft.

[pixel_popping]

A script made by AI, digested by AI, organized by AI? Let's call it what it is.