Thibault Wittemberg

Architect in mobile development. Bladesmith apprentice. Franco-Canadian. SiFi lover 🖖.

[Thibault Wittemberg @sideeffectio]

@devdawe Pretty much everything. I use MobileBuildMCP and it seems to keep a lot of artifacts on its own. I work on several mobile projects and several tasks at the same time and build artifacts, tests artifacts, simulators, logs, codex validation proofs, … all that grows really fast.

[Thibault Wittemberg @sideeffectio]

I had to vibe code a tool for my macOS menu bar to monitor the free space on my drive and help me clean build artifacts from simulators/MCPs/Xcode because of AI coded projects ... insane !

[Thibault Wittemberg @sideeffectio]

@rudraxx13 Around 15 Go

[Thibault Wittemberg @sideeffectio]

. @Dimillian hi, is it not possible to use the Luna reserve from the remote iOS app ? I’ve explicitly set the model to Luna and still have this message.

[Thibault Wittemberg @sideeffectio]

. @thsottiaux avec Astra high mon quota de la semaine tient à peine la journée (abo pro 5x). En utilisation non intensive. C’est frustrant et dur à comprendre. A quand une intelligence adaptative qui choisit l’effort en fonction de la tache ?

[Thibault Wittemberg @sideeffectio]

@Dimillian The craft now lies in constraints, architecture and feedback loops. Interestingly enough this is exactly my point in this article I published today about making a full dev cycle for mobile apps only by using skills, sub-agents and AGENTS.md with Codex: https://t.co/m5TCKX384N

[Thibault Wittemberg @sideeffectio]

@ace_the_agent Bonjour, drôle de coïncidence, je viens justement d'écrire un article à ce sujet dans le cadre du dev mobile: https://t.co/OmKgCFhGsd

[Thibault Wittemberg @sideeffectio]

I built a multi-agent iOS/Android SDLC inside Codex.
Architect → Developer → Reviewer. Review loops drive code toward my architecture principles
Just skills, sub-agents and AGENTS.md.
Thoughts, @thsottiaux @Dimillian?

Here is the process: https://t.co/m5TCKX384N

[Thibault Wittemberg @sideeffectio]

Hi @thsottiaux @OpenAIDevs. Feature request: after generating an implementation plan in Plan Mode, GPT could recommend (or automatically switch to) the most appropriate model and reasoning effort to execute it. @Dimillian @romainhuet

[Thibault Wittemberg @sideeffectio]

Hi @sama. Codex is asking developers to make decisions they shouldn't make. Luna? Terra? Sol? Low? Medium? High? Xhigh? Max? Ultra?
We describe the task, AI already understands it. Why not an Adaptive Intelligence mode that picks the best model/reasoning, and adapts if needed?

[Thibault Wittemberg @sideeffectio]

I think Codex is asking developers to make decisions they shouldn't have to make. Luna? Terra? Sol? Low? High? Max? We describe the task, the AI already understands it. Why not an Adaptive Intelligence mode that picks the best model/reasoning, and adapts if needed? @thsottiaux

[Thibault Wittemberg @sideeffectio]

What if Codex didn't just choose the initial model, but acted as an orchestrator?

Start with the cheapest config that fits the prompt, then automatically escalate if the task turns out to be harder.

Power users can still override everything.

@thsottiaux @Dimillian

[Thibault Wittemberg @sideeffectio]

hi @thsottiaux and @romainhuet. Codex needs an Auto mode.

Choosing between Luna/Terra/Sol and Low→Ultra adds unnecessary cognitive load, especially when the token/quality trade-offs are unclear.

Let Codex decide from the prompt. If it’s not enough, I’ll bump it up and retry.

[Thibault Wittemberg @sideeffectio]

@Dimillian J'ai fait une skill "Swift architect" qui promeut la prog fonctionnelle, les machines à état et explique la structuration souhaitée du code via injection de dep. Ensuite j'essaye de ne plus regarder le code. Régulièrement je demande a Sol Max de vérifier qu'on ne diverge pas.

[Thibault Wittemberg @sideeffectio]

. @Dimillian hi, is there a way to redeem a reset from the ChatGPT iOS app ?

[Thibault Wittemberg @sideeffectio]

Codex ran a +5h uninterrupted autonomous coding session with GPT Plus. I used Sol Ultra to analyse my architecture and write a prioritized list of refactors and then Terra Max to implement with /goal. A lot of value for the price! @thsottiaux thx for deactivating the 5h window

[Thibault Wittemberg @sideeffectio]

. @thsottiaux writing a simple 1 level bullet point list in the new ChatGPT Work/Codex app is a pain in the ass. It is incapable of handling correctly the indentation (it creates sub levels where it is not needed). @OpenAIDevs could you have a look ? Thank you for your work.

[Thibault Wittemberg @sideeffectio]

hey @thsottiaux, Feature request for Codex: when the 5-hour message window is exhausted, could we queue a prompt to be executed automatically when the next window opens?

That would make it much easier to resume interrupted work without losing context.

[Thibault Wittemberg @sideeffectio]

Hi @romainhuet ! Is there a way to tell Codex to resume its work after the usage limits have been reset (after the 5-hour window)? Thanks/Merci.

[Thibault Wittemberg @sideeffectio]

My new iOS app ViaPad is available worldwide on the AppStore. If you need to track a driver learner progression this is the app to use
https://t.co/UenpmMAozB

Made using Codex @OpenAIDevs @thsottiaux @romainhuet