Mattia Cerutti

first time shipping something like this i built a small tool for big tech interview prep, mainly because i hate the whole leetcode thing and was forgetting half of what i solved it basically sits on top of problems and keeps track of what you did, what you should review, and what to do next still rough around the edges, but it’s been useful for me so far. If you’re preparing for interviews, curious to hear what you think: interviewprep.sh

@b_nnett Codex App is not oss, right? How does this work?

Edit files directly within Codex, now available.

@picoito I'm not sure that would go well with the current design capabilities of GPT models 😅

i kinda wish codex had a design module, like claude

@4ster_light Just go for Go, insane value for 5$

“Switch to 5.3 Codex” YOU REMOVED THAT FOR STUDENTS TOO? Seriously I should get OpenCode Go, I just don’t have the time with finals now

@pcshipp Why ads and not UGC? Genuinely curious 👀

Hey devs, I’m ready to invest $500 in ads. Which platform is best for ROI? - Meta ads - Apple ads - Google ads - Tik Tok ads

@ItsAlexhere0 If that happens, i think the majority will shift to product building. Everyone will basically have its own startup with AI employees

Software engineers, what’s your plan B if AI writes better code than you in the next 3 years?

figure out a way to have a working undo/rewind feature in both the codex app and CLI really can’t understand how people can work w/o that. Being able to undo my work through a checkpoints system is one of the most basic features a coding agent should have

@talhasarit Yes exactly! But eventually i think they will, and that’s when we won’t be micromanaging them anymore. Until then tho, i still don’t trust them that much

@mttcer Agreed. AI agents still don’t have “taste” . They will almost always generate slop whenever you let it work autonomously with no steering.

not a critic to Max, but i still don’t get how people can let agents run this long. for me, i use agents only for planning / coding individual features, but i still need control over the overall codebase and decisions i feel like that if i let an agent run for hours i’ll come back with a messy codebase that would take days to understand and review

@paulhshort happy that it’s working out for you. I feel like those things you mentioned help the agents reach the end goal with a working build, but don’t really help in terms of security, performance or a clean implementation.

@mttcer This is nothing, I've had it go for 8 hours on several occasions, you just need to close the loop and give it ways to validate and test its work

@andreisavin I am doing this to some extent, it’s more of a trust issue. I know it makes you more productive by a ton, but personally i think llms are not in a state where i can fully trust them to produce good and safe code without supervision yet

@mttcer you need strong guardrails to let long-running agents loose in your codebase: enforce testing, coding style, have another agent validate before committing, clean up periodically (running claude's /simplify command is great)

@thatdna Cause when something breaks, i should be able to fix it.

@mttcer Why on earth would I need to understand the codebase

@walidbelr I don’t think we’re quite there yet. Yes, llms are definitely exponentially improving, but they can still fail, produce unsafe or non-performant code. Software quality is decreasing, we have more outages and security incidents. At this moment, we cannot blindly trust AI yet.

@mttcer The world has changed, my friend. You don’t need to know every single line of code anymore. But you do need to automate tests so you can be sure old features aren’t broken. Either you adapt and adopt this, or you’ll be replaced. Hard truth, sorry.

@OpenAI Curious about what are these other local agents supported 👀

Bring your workflow to Codex in just a few clicks. Import settings, plugins, agents, project configuration, and more so you can keep working with fewer interruptions. Your move.

@RhysSullivan unfortunately it’s often the other way around 😔

my system handles an edge case i didn't account for

@nummanali I still don’t get this everything app though. Is it going to consume only codex-related limits like now? Because that would mean wasting all the chatgpt messages that come along your subscription

I realised why there isn’t a Codex app on mobile It’s very likely the full ChatGPT brand will phase out in favour Codex When Codex is being called the everything app, it makes complete sense So they’re probably setting up the right infra for the 900m+ user transfer

@noahreeveshq here!

Who the hell is using VSCode? 🤔

@devbelowstairs yeah definitely unfollowing after this switch

I have changed my profile picture from this cute looking robot to this ugly looking guy, why you may ask? pros: - so people know im not a bot and a real person - this grey filter makes me look like one of those "tech bro" profiles - cant think of anything else cons: - the only thing interesting about my acc is gone - no more cute bot profile pic - its a downgrade - might scare away people

@daradoescode @t3dotcodes only thing that’s still keeping me from switching to t3code is the fact that skills are tied to the provider you’re using. i know that’s kinda obvious since it’s using the provider’s CLI under the hood, but i really wish there was a solution to this

the codex team is shipping a lot of cool stuff but the performance of the app kinda blows i feel like I always have to work in either @t3dotcodes or dpcodes n then whenever I need to do stuff like image gen etc... I gotta move over to codex