Jasmine Damani

@webdevour.bsky.social

Software nerd, cat enthusiast, pixel art noob, gamedev, playing with LLMs

Hobby project of mine in production is apparently using github oauth secret from a github account that I don't remember setting up and likely will never be able to sign in to again... 🤣

I used to hate shipping new features. One bad deploy = rollback stress + support tickets. Then we started using feature flags the right way. Here’s the exact approach that let us move much faster (and sleep better) while building Basestack 🧵

Correct me if I'm wrong, but doesn't the notion of temperature in LLMs mean that some small % of the time your "coding agent" might decide to do something crazy? Like ignore all the guidelines you packed into the context? Or do "thinking models" make it probabilistically impossible?

One thing I'm not sure of is whether writing up a big cursor rule or skills file for some section of my code is worth the time? It seems possible that the LLM is better at inferring what the architecture is and the various quirks and even UX intent is without me writing it up in prose.