Raul Saavedra

Software Engineer 🖥️ I post on about software, LLMs, and the apps I’m building. Currently running @openai Codex

[Geoffrey Litt @geoffreylitt]

Hot take: I think it's still important to understand the code that our agents write!

In this mega thread (based on my AIE talk today), I will explain why that's the case, and show some ideas for how to efficiently understand code. Alright, let's dive in. 1/

[Raul Saavedra @raulsaavedra_]

@cpjet64 On my end it’s taking almost 15 minutes to compact on gpt 5.4 high 😅
Must be a bug @thsottiaux ?

[Raul Saavedra @raulsaavedra_]

Codex randomly cussed mid-session without me prompting it. Haven’t seen that before.

[Raul Saavedra @raulsaavedra_]

Love when Codex adds a fallback for a fallback for a fallback.

[Raul Saavedra @raulsaavedra_]

@tritlo TUIs just feel so much faster than using GUIs for me. I like that they can live in my tmux sessions and I can multitask more efficiently 😅

[Raul Saavedra @raulsaavedra_]

@DavidKPiano @RhysSullivan “Hey GPT, can you turn this simple event handler into an useEffect?” 😂

[jason @jxnlco]

Send me your non coding Codex demos!

- 30-50 second video

music production, game development, 3d, automations, anything. anything good we'll shoot over some credits so you can go deeper!

[Raul Saavedra @raulsaavedra_]

@jxnlco Probably counts as coding, but built a skill-tree TUI that comes bundled with a Codex skill. It has been pretty useful for learning new stuff!
https://t.co/7BWOoFoCW5

[Raul Saavedra @raulsaavedra_]

@0xJuicy__ thanks bro! honestly not right now, but very open to any feature requests on the repo!

[Raul Saavedra @raulsaavedra_]

Fully open source. Open to any feedback/new features!
https://t.co/ccEbm5eQBa

[Raul Saavedra @raulsaavedra_]

I have been using it to refine my AWS fundamentals.

My favorite part is that I can ask questions based on the quizzes the agent generated.

If I want to go deeper on one of the quiz questions, I can ask right there and the quiz becomes part of the lesson.

[Raul Saavedra @raulsaavedra_]

@jxnlco For real! I was using Opus because I couldn’t stand 5.3 codex’s personality. Tried 5.4 and switched immediately 💀

[Boris Cherny @bcherny]

Introducing: built-in git worktree support for Claude Code

Now, agents can run in parallel without interfering with one other. Each agent gets its own worktree and can work independently.

The Claude Code Desktop app has had built-in support for worktrees for a while, and now we're bringing it to CLI too.

Learn more about worktrees: https://t.co/JFkD2DrAmT

[Raul Saavedra @raulsaavedra_]

@thsottiaux Hooks feature would be great! It would allow use cases like running custom scripts on session start, attaching Codex sessions to tmux panes, or notifications

[dax @thdxr]

i use opencode a ton - it types out all the code for me

but i know how everything in my codebase works

anything that's subpar i'm aware of and i have a sense of how to fix it eventually

when ai fixes something i don't move on until i know it's the best possible fix

anything that requires hacky fixes i slow down and re-evaluate the whole system

[Raul Saavedra @raulsaavedra_]

For me it seems a more broad knowledge model like GPT 5.2 might be better for chat, but Codex is known to be better at tool calling and running coding tasks for a long time, so I am not 100% sure

[Raul Saavedra @raulsaavedra_]

For those using their Codex Auth with @openclaw, which do you prefer, GPT 5.2 or GPT 5.2 Codex?

[Geoffrey Litt @geoffreylitt]

Unironically the best AI coding trick I’ve found recently: forcing myself to pass a quiz on how it works before approving agent code 😂

The illusion of false understanding shatters fast

[Raul Saavedra @raulsaavedra_]

@steipete Yeah even though I want to automate as much as possible with Codex or GPT, it’s just not a good idea for people to people interactions. It ends up looking silly 😅

[Raul Saavedra @raulsaavedra_]

@steipete How did you get it to speak like that? Mine has no soul 😂