Rahul Patil @RahulScripted

@reve can you please expose this in API have a very big use case

@skirano Hey! Just wanted to check if there is something that can break an AI image into editable elements? like Canvas Magic Layers. But exposed via API? Have a very big use case but unfortunately no such tool via API exists

@leojrr Hey! Can I get some credits to try it out

@sama I am working on trying to decompose an AI generated image into editable components. Somewhat like the Magic Layers in Canva. But more accurate. Also Canva doesnt expose the Magic Layers via API. Would love to connect or get help on this

@oliverkenyon is there a canva magic layer api?

For context — this was on a real task: DB → JSON → spatial geometry analysis (x/y) → overlap detection. Codex handled it cleanly in one prompt on the $20 plan. Claude (on the $200 max tier) kept looping—and at one point even said that It is correct and that I might be wrong.

@thsottiaux Loved the updates you’re rolling out to Codex. Ran a few fairly complex tasks across both Codex and Claude interesting outcome. Codex handled them cleanly in a single prompt, while Claude kept looping even with manual interventions. Safe to say we’re making the switch

Hi @lovable , I dont want to use your Lovable Cloud, I want all my tables and backend to be connected to Supabase, however the app doesn't allow me to? Worst of all your lovable cloud doesnt allow me to connect to the database? I dont see any connection details

@ForrestPKnight More than 90% at this point! Not sure if this is right, but at this point I ask the agent to code and instead of checking each and every line of code, I just check if the feature is working, and if it works it works! However I have added a lot of guard rails on the agent

@budapp Hey cant wait to test it out! we currently use Claude as backend engineer and Lovable as our front end

@sama to try it right away, make sure to first update codex cli to v0.36.0: `npm install -g @openai/codex@latest` and then run codex with: `codex -m gpt-5-codex -c model_reasoning_effort="high"`

Is it me or the new OpenAI responses api is kind of similar to the apple ecosystem. The new API is soo good for building tools and agents but the catch is that you can use only OpenAI LLM's. Fair on them to try to push and make their product better but you get locked in to OpenAI

I have to build an agentic workflow for a product. Its main task is based on users prompt it should choose from a bunch of tools available to it and run them in a particular order decided by an LLM. Split between using the OpenAI Responses API vs Langchain. Any recommendations?

@GeminiApp @Google #Gemini #GoogleAI

Was trying the Gemini Banana Model. It looked so promising at first for image editing. But when asked to remove a simple shadow from below the car it failed! Tried this was all prompting strategies but I simply cant get it to work! Is it just me or am I doing something wrong?

the best part! the repo at a bird eye view conneected to a data source and published it to various consumers. Claude asked me at the end if I want simulate the remote data source locally for testing by creating a config to redirect the connections

I primarily code in python everyday however today I had to work on a hefty GoLang Repository. I have 0 knowledge of golang and have never worked with it before. I spent about 10-15 minutes trying to understand the codebase and even though I understood the overall process.. (1/N)

the documented it crated was so thorough! ait was a like getting a KT from a senior developer regarding the repository It had everything from how to install with the exact version followed by documentation for each api along with its parameters and data types

It was time consuming for me to understand the details of API such as the allowed params. authentication the detailed conditions and fallbacks for each API. So I asked claude to generate a detailed readme for the repo with steps to install and a detailed doc for each API used

One thing that I found useful with claude is to give it a feature and ask it create a detailed implementation plan in a "planning.md" file. Then as a follow up prompt ask it to refer the planning file and implement all the steps. Makes fewer errors!