Xbotter

AI Accelerator × AI-Native Builder
Helping enterprises land AI successfully.
Code w/ 🐱

[Xbotter @Xbotter]

A brief formal introduction:

With over 10 years in .NET development, I’ve fully shifted my focus to AI-native applications and enterprise AI workflow implementation.

I specialize in turning LLMs, Agents into practical solutions with real business impact.

Lately, I’ve been exploring tools like codex in depth.

Outside of work, I’m a dedicated cat dad 🐱

Happy to connect on .NET + AI + codex topics.

[Xbotter @Xbotter]

@thsottiaux @jxnlco May the tokens be with you.

[Tibo @thsottiaux]

Hi. Over the last 24 hours we had three separate small incidents that affected Codex reliability. Those are three too many and we are taking active steps for them to not reproduce.

I have reset usage limits for Codex across all paid plans. May the tokens flow again.

[Xbotter @Xbotter]

@thsottiaux If this short film was created with Codex, it would be absolutely perfect.

[Xbotter @Xbotter]

@thsottiaux It currently seems like multi-threading is restricted.

If slow it down, single-threading can still run, but my productivity has taken a huge hit.

[Xbotter @Xbotter]

"exceeded retry limit, last status: 429 Too Many Requests, request id: xxxxx"

It looks like there are indeed some issues with the codex usage limit.

@thsottiaux is probably already getting ready for the reset, right?

[Xbotter @Xbotter]

@thsottiaux @romainhuet "The model 'gpt-image-2' does not exist." ?

[Xbotter @Xbotter]

OpenAI just showed how to build self-improving agents with Codex.

Not a one-time deployment, but a system that keeps improving in real production environments.

This framework is worth studying for any team building agents.

🔗 https://t.co/RE2bQjszyu

The biggest problem with traditional agents: after launch, you keep discovering edge cases.

Fixing them usually means engineers manually tweaking prompts and digging through logs — slow, painful, and entirely dependent on humans.

This project creates a true closed-loop self-improvement cycle:

Production usage → Structured signals → Codex-driven iteration.

The loop is built on three key pillars:

Close to practitioners
The people who actually do the work steer the system. Their corrections determine what matters most.

Production traces as evidence
Capture the full path: raw documents → extraction → mapping → expert corrections → final output.

Codex-driven iteration
Turn expert fixes into structured findings, package them into targeted evals, then let Codex investigate root causes, edit code, run validations, and submit PRs.

Clear patterns are automatically optimized. Ambiguous ones go back to humans.

That’s the loop.

Real-world example:

Experts keep correcting the same field → the system detects the recurring failure → automatically builds an eval dataset → Codex analyzes the pipeline, extends the schema, updates the mapper, validates the change, and submits a PR.

Every improvement generates fresh production data that fuels the next cycle.

Within just six weeks of launch, accuracy improved dramatically, and the system was able to handle far more complex tasks efficiently.

Why this framework is powerful:

It works in any expert-knowledge domain: accounting, law, operations, customer support, healthcare, and more.

Engineers focus on architecture and final oversight, while Codex handles routine improvements.

Domain experts naturally provide high-quality signals through their daily work.

Self-improving agents are a practical path you can start building today.

[Xbotter @Xbotter]

@Dimillian /side then prompt

[Xbotter @Xbotter]

@thsottiaux Better late than never — have a wonderful weekend!

[Xbotter @Xbotter]

@thsottiaux I looked at a number today on a codex dashboard and it made me sad.

Usage remaining 0%.

[Xbotter @Xbotter]

The improvement in intelligence comes with fewer guesses and more effective goals. This is not merely a best practice in model definition, but something that allows for a much better alignment with the user’s personality.

This is not an effect that a model alone can achieve — it is the perfect synergy between the application and the model.

[Xbotter @Xbotter]

@harjtaggar Spot on. With these natural resistors, skip “how to start” advice. Drag them into worst-case scenarios instead. Their instinctive pushback = proof it’s effective.

[Xbotter @Xbotter]

@thsottiaux What else can't codex do?

[Xbotter @Xbotter]

One day, you’ll be able to do anything in Codex without ever leaving the chat box.

[Guinness Chen @guinnesschen]

If you ever get tired of managing your Codex threads, just let Codex manage itself! Codex can now create threads, search them, organize them, pin the important ones, and spin up worktrees for parallel tasks.

[Xbotter @Xbotter]

@guinnesschen One step closer to self-reflection.

[Xbotter @Xbotter]

A simple trick:
If you use the Codex app linked with GitHub code review, you can have Codex submit a PR and at the same time set a scheduled task (every 10 minutes, depending on the size of changes).

The task tells Codex to check the review comments, fix them, commit and push the fixes, and repeat until there are no new review comments left. Then delete the scheduled task.

The Ralph loop for cloud collaboration.

[Xbotter @Xbotter]

@thsottiaux Codex (formerly ChatGPT). Is this the dream?

[Xbotter @Xbotter]

@thsottiaux /fast

[Xbotter @Xbotter]

I absolutely love this cover image. The distant light guiding a lonely soul — is it humans who discovered the light, or the light that guides humans?

People who can truly see the future are incredibly rare. So much of what people say and so many so-called truths are just hearsay. Only those who have truly seen it for themselves can offer something beyond empty opinions or secondhand stories — they share insights from a path they have genuinely walked or witnessed.

I believe YC must have accumulated a wealth of such exploration stories. These aren’t the kind you can read in business magazines or hear in novels. Success stories may look glamorous on the surface, but only a few people ever get to see the immense effort behind them.