Sarthak Langde

engg @ building

[Sarthak Langde @sarthaklangde]

https://t.co/ZU9RhRpioy

[Sarthak Langde @sarthaklangde]

@marckohlbrugge man, coincidentally barcelona has burtal competition of voice ai companies. we have built voice calling agents to bring more people into retail clinics as a service and honestly, if it's someone's AI talking to our AI, we would be so happy. such efficiency.

[Sarthak Langde @sarthaklangde]

you are telling me that you can’t be bothered to create a new row in DB? or even worse, that you don’t have an AI agent that automatically takes care of this?

[Sarthak Langde @sarthaklangde]

be me (and my team)
win $30k in OpenAI credits for winning hackathon
need registered OAI org to receive credits
ok
teammate requests additional org
OAI team assigns a random org with other users in it
teammate asks OAI team what to do about those users?
OAI team: “just remove them”

[Sarthak Langde @sarthaklangde]

bruh we need a pre-commit skill for every member on my team. your claude cannot be writing docs in a different way than my codex

[Sarthak Langde @sarthaklangde]

@OpenAI @ChatGPTapp desktop app is broken on Mac. I can't scroll through the chats. It is as if the overflow field has been disabled or broken.

Works fine on web, but please fix this.

[Sarthak Langde @sarthaklangde]

Codex agent can do all my work while I spend countless hours pondering over the purchase of a single-use $20 gadget

[Sarthak Langde @sarthaklangde]

Bye-bye Reddit rabbit holes.

DeepSearch mode on Grok has replaced my habit of analyzing public opinion on products with reddit.

Instead of "whoop mg reddit", I can now just tell Grok to tell analyze articles + public sentiment, and then also give me the best way to use it.

[Sarthak Langde @sarthaklangde]

i am done creating crawlers that work only once

every useful system i have built relies on continuous data sync

so can i actually use AI agents to make crawlers that self-heal, maintain and deploy themselves?

i am gonna try

[Sarthak Langde @sarthaklangde]

speedrunning through 7 stages of grief over a nearly abandoned SaaS app

AI empowers, but no meaning left in the journey of coding the app itself anymore

finding it super hard to commit to any idea when AI can build any new idea in 5 min

need that sunk cost fallacy somehow

[Sarthak Langde @sarthaklangde]

@lovable 6/ Bonus from co-worker

"What did @lovable say to https://t.co/7VgE0Eov70?"

"skill issue"

[Sarthak Langde @sarthaklangde]

I tried to generate a simple background image removal app using https://t.co/q3s0wpmXKP and @lovable.

My prompt?
"Make a Next.js app with Tailwind, TypeScript for background image removal.

I want the user to be able to signin with Oauth. The homepage should have an upload button where the user can submit the image and click process. If they aren't signed in, they should be redirected to the sign-in page.

The backend route should be /api/process-image and it should take the image and return the processed image. Use simple npm package for background removal."

🧵

[Sarthak Langde @sarthaklangde]

@lovable 5/ @lovable actually created an app that worked for background removal without much fuss 🎉

PS: the background removal itself isn't that great but that's cause of the simple npm package being used.

[Sarthak Langde @sarthaklangde]

@elizabet_dev Youtube Music user here. 0 hope of finding anyone else using it.

[Sarthak Langde @sarthaklangde]

@runpod quick deploy with VLLM (serverless) is broken. It doesn’t matter which model you choose to deploy, it keeps deploying facebook/opt-125m…

The MODEL_PATH env variable is not being passed correctly. Its been 20hr+ and this issue still persists.

[Sarthak Langde @sarthaklangde]

This thread was not inspired at all because of my messy Downloads folder that could be cleaned easily by LLM automation. Not at all. Really.

[Sarthak Langde @sarthaklangde]

How can we use system AI across apps?

Just like ARIA labels for web, we should have LLM labels for web, Android and iOS for buttons. This would allow developers to specify which elements on the app/page can be controlled by LLM while preventing UNSAFE actions.

[Sarthak Langde @sarthaklangde]

4. All apps are going to be LLM controllable sooner or later. As AI builders, we need Apple and Google to quickly address this gap and include platform-specific instructions in the SDK ASAP.