Rami M@rami_mithalouniMay 5, 2025
I did a little Saturday morning vibe coding to cook up some fun images for the Voice AI Course that we're doing this month.
We'll comp the course fee for the first person to tag all the course guest speakers who are in this image. :-)
Repo link and some notes about this particular vibe coding experience in the 🧵
Source code for the image manipulation scripts that Windsurf and Claude Sonnet 3.7 wrote: https://t.co/dRDm7F8Ic8
Link to sign up for the course ($10k in free credits, 28 sessions with Voice AI experts): https://t.co/yjxxM3Brtw
Windsurf + Claude did a really good job writing this code. (More notes about that in the README file in the repo.)
But the big thing I came away thinking about on Saturday morning, though, wasn't LLM code gen capabilities. It was that this shouldn't have been a programming project at all.
I can think of two much more natural ways to do this kind of image manipulation.
One, I could have just asked Photoshop to do this for me. "Hey, Photoshop, here's a workflow I'm imagining. Do it once. Maybe we'll need to iterate together on it. Once I like the result, we'll do it a bunch of times and save it as something we can come back to later."
Or, I could just ask the Universal LLM Chat App That We May Soon Do All Our Work Inside to do this for me. The LLM can write code. The LLM can manipulate filesystem-like entities. The LLM can call out to tools. The inference environment or the application environment or both can include scaffolding that runs code.
I don't ever have to see the code. I mean, I personally want to see the code, maybe. I like code. But maybe even I don't care that much about the code, for something like this. And certainly most people won't.
@rami_mithalouni won the contest!
https://t.co/BUvXhtcqKz
Rami M@rami_mithalouni@kwindla Excited for the course! here's everyone :)
@Prince_Canuma @DynamicWebPaige @isidentical @philip_kiely @natrugrats @rajivayyangar @mark_backman @shresbm @kwindla @swyx @aconchillo @charles_irl @lilyjclifford @zachk @qfav95 @qfav95 @bnicholehopkins 1/2