𝖕𝖗𝖆𝖙𝖍

ai vamp | sde | cursormaxxing

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

I just made a short video showing how to do this with the CLIProxyAPI tool. It's a very useful and easy-to-use tool, but it can be hard for people who have never used it before. In this video, I showed you how to use any OAuth subscription in Cursor AI.

Links below ⬇️

[Fekri @fekdaoui]

hey @cursor_ai i’d love to be able to use my openai codex sub in cursor’s native agent window

you could even make it available only to active cursor subscribers so people don’t just free-ride

cursor ux + codex sub usage limits would be elite

[parth. @TotallyNotParth]

Tested several open models on their frontend capabilities: same prompt, same harness, same infrastructure... surprised with the results! Would like to know your thoughts. Big thanks to the @gmi_cloud team for the credits! 🧵

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

I'm just uninstalling all agents'/AI-related coding stuff from my laptop and going to do a fresh start with Cursor (Cursor subscription) and OpenCode (OpenCode Go subscription) and stick to them.

As for skills, it's only mine and Matt's; those are all I need for day-to-day work.

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@ludvigrask_ @opencode amazing work

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@hishadman It was: Use @Browser and verify the UI/UX ans fix any issues you spot

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

Anthropic literally cooked. Claude Opus 4.8 is amazing.

Find the whole design in comments.

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@shamikv @biraj21_ @amay_bakshi 2. The interviewee passed the interview presented for the pay. Isn't that enough to qualify the person? Or should the interviewers conduct easier/harder interviews for less/more pay based on the last CTC?

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@windsurf y'all cooked, antigravity was already bad, this just killed it lol. only ai team i'd trust apart from cursor is the devin team 🫡

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

> found a new tool for codebase sanitization
> too lazy to run it locally and watch over it
> go to claude and make it write a long recurisve prompt
> put it in .@cursor_ai cloud agent with grind mode on
> plans it start to end, i approve

mfw when i realise i'm addicted

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@CircuitSage_ yeah, i need to try it out and confirm

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

has anyone built cursor cloud agents like ui/ux but self hostable? also works with mobile?

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@traviscurnutte @cursor_ai agreed lol

[siddharth @buildwithsid]

me in 2013 instead of doing Leetcode

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@vibercoding true

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

trying opus 4.8 + minimax m3 combo in claude code (opusplan mode) and seeing m3 error out a lot in cc, where as it worked just fine in opencode and pi

weird

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@SantoshYadavDev very inspiring story 🫡

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@CircuitSage_

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@DOdeniyi9389 ok, i'm back after trying it a bit more, i agree with you

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

it'll be interesting to see how m3 ranks on this because m2.7 just straight up failed, and that's accurate to how it felt in real world usage too.

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@wardies @skcd42 ahh i wanted to try without buying a grok sub 😅

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@bridgebench agreed. to truly use it effectively you need to pair with either opus 4.8 or gpt 5.5 and make them plan the whole thing out start to end and then execute with m3.

that's what i did here and got better results:
https://t.co/ZTTxHgwlpY

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

like here's an exploration:

natural text to resume with editable fields

opus 4.8 did all the planning, and m3 wrote every single line of code; none were written by opus, only planning.

explaination below:

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

@melvynx m2.7 was genuinely bad, i agree. but man m3 is A LOT better i swear...

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

overall, m3 is an excellent worker model, like composer 2.5 is, but it needs a co-ordinator who can take the user inputs and transform them into detailed specs...

that's all! thanks for reading (if you did)

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

like here's an exploration:

natural text to resume with editable fields

opus 4.8 did all the planning, and m3 wrote every single line of code; none were written by opus, only planning.

explaination below:

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

ngl minimax m3 is a nice model but as a worker, on its own i found it quite weak in terms of agentic use. I’d describe a bug lightly and it’d make a bad fix, where as pair it with opus and have opus write a plan for the fix and then m3 executes it perfectly

[𝖕𝖗𝖆𝖙𝖍 @prathamdby]

re-explained the same thing with same prompt and element from react grab, opus understood immediately made a plan, executed via m3 and voila bug fixed, repeated that few more times and got what you see in the video. i didn't make it integrate ai fully because i didn't have time