Kien Nguyen

[Kien Nguyen @kiennt2288]

@ValsAI @grok what is Lean in here

[Kien Nguyen @kiennt2288]

@ffsxdev How did you get game state from the app? and interact with the app as well?

[Kien Nguyen @kiennt2288]

@Richelle_Ji @taurobots @corneliawein @UnitreeRobotics Are you sure the robot clean its hand before wrap your clothes?

[Kien Nguyen @kiennt2288]

@thsottiaux I think ChatGPT Desktop improve very fast. It also allow me to control from mobile, especially start a new session from mobile while Claude does not allow to do.

[Kien Nguyen @kiennt2288]

@thsottiaux For me the main feature is because new models like GPT 5.6 Sol and Luna are both good.
Sol is good for plan task, and once I have plan, Luna is quite cheap to run.
Also, I kind of GUI person who really enjoy to work in ChatGPT Desktop Codex: show list of agent, side chat, ...

[Kien Nguyen @kiennt2288]

I think ChatGPT will win over Claude Code. Fable still smarter than GPT 5.6 Sol, but the gap is very small now.
ChatGPT catch up so quick, new models like Sol, Terra, Luna is smart enough for most of my works.
And the desktop app is much better than Claude Desktop.

[Kien Nguyen @kiennt2288]

@aryanXmahajan OS

[Kien Nguyen @kiennt2288]

@OpenAIDevs I done a similar thing, and allow user to point at any element on the simulator, drop a comment, and Codex picks it up as context

Integrate with OpenAI Codex is shocking easy.

[Kien Nguyen @kiennt2288]

@Baconbrix @cursor_ai I done a similar thing. And allow users to point at any element on the sim, drop a comment, and then send the comment to the agent

[Kien Nguyen @kiennt2288]

Inspired by @Baconbrix, I extended Codex with an iOS-simulator tool this afternoon — point at any element on the sim, drop a comment, and Codex picks it up as context. Codex is shockingly easy to extend @gdb

[Kien Nguyen @kiennt2288]

@Baconbrix I am curious about how could you make the interaction between the simulator on the web so smooth.

I guess to get the preview of apps on web, you have to make a streaming server for simulator. But it will make the significant delay for interaction.

[Kien Nguyen @kiennt2288]

@charlieholtz Next could be allow people to chat with conductor from slack / tele or any messaging app

[Kien Nguyen @kiennt2288]

@_can1357 I wonder why your hashline need hash instead of raw line number? What is purpose of hash in here?

[Kien Nguyen @kiennt2288]

@mattpocockuk This pattern works very stable. I could make a plan and tasks to run for 6 hours

[Kien Nguyen @kiennt2288]

@mattpocockuk I tried to use stop hooks like that, but it does not work very well.
After that, i used this pattern:
1/ Ask agent to create a tasks file with detail for each task, including test cases and how to verify
2/ Ask agent to spawn sub agent foreground, each agent only do a task.

[Kien Nguyen @kiennt2288]

@hthieblot could you share your framework of decision to fund for a founder?

[Kien Nguyen @kiennt2288]

@OfficialLoganK @GoogleAIStudio would love to try it out, can wait to replace all my tool to Google and Gemini.

[Kien Nguyen @kiennt2288]

@kat_kampf @GoogleAIStudio Would love to test it. I got more excited with aistudio than other tool like Loveable and v0

[Kien Nguyen @kiennt2288]

@donvito So the question should be like how you could utilize it. For daily coding you may not reach the limit. But if you want to run autonomous agent, it is worth it.

[Kien Nguyen @kiennt2288]

@donvito It really depend how you could utilize it. I use ccusage to monitor my usage. Claude Code is limit by week. I use 200$ MAX plan, and in 3 days, I could use ~350$ token ~ 37% weekly usage. So it could be ~ 900-1000$/week.
If you use 100$ MAX plan, it could be 220-250$/week