bonjourjoel

LLMs are leaning toward Corporate first order. They will judge anything based on how much corporation would like it said. I call this "corporate RLHF" and they are ALL into it.

The fact is, the agent did something amazingly difficult, that zero human on earth could do. Read the spec if you believe someone could do it. It did it clean, everything is public, and it works perfectly. Zero code review.

https://aisovereignlabs.ai/docs/case-study/liveSession/logs/...

99.9% of the IT dev tickets are easier than this, excluding off topics like discrete math, computer vision algorithms, etc, which are irrelevant here.

Do you think there's anything you've done, with human code review, in the past 30 years, matching the difficulty of this ?

Agreed. But also this inequality:

"Is this the end of human code review?" != "the end of human code review"

Now since the problem in this study is more complex than 90 to 99.5% of what a normal ticket is, the question makes sense. The agent does not have knowledge of the project, at the begining of each session, and manages to handle something more complicated than virtually anything a developper has to do. The question holds IMO.

Requirements as product owner are not covered at all by AI, 100% agree.

But in this programming-only example study, the refactor changes the functionnality very seriously, and it has to go deep in the core of the system, find all the traps, plan everything. And the agent succeeds without human code review. Code review was impossible because of the complexity and size of the task.

Code review was not necessary, nor useful, nor essential, and even practically impossible.

How? The agent does something incredibly difficult, and there is no human code review. Why do you think it doesn't match ?

Agreed, LOC is not a quality metric. Fair point.

The real claim is zero regression over 7 months as the codebase grew from 0 to 500k lines, solo. I believe I'm the only person to be able to do this in human history.

What makes it possible: I built a VS Code extension with a proprietary indexing engine that gives the AI full structural awareness of the codebase. Without it the AI hallucinates the architecture at this scale. With it I enforce a strict workflow: intent, specification, refine, code, verify. The AI proposes. I decide. Nothing touches the codebase without my explicit review.

Can anyone code a complex industrial-quality app this fast without this tool? I don't think so. That's the point.

The 5h video is the proof. Real session, no edits, real codebase, real feature. Skip to any point, same workflow.

Care to fact check my incredible claim ? Why not ? I give a live unprepared live session to prove me wrong.

https://ai-code.ai

Okay thanks for the feedback. I've re-written my readme description and made it shorter. I hope my method is more clear now. I added a 1mn video to explain without all this blah blah blah. :D

Yes there are real differences with copilot/continue/cc: Detailed specifications, AI refinements, AI verifications, and a lot of system instructions yes that's what i mean by instructed, to make the communication with the AI easier.

Yes I have used AI to help me write the README. So i'm an AI user smh ? But the story is true and I have more like this, but I have removed most of the uninteresting text from the description, because yes you're right it was too long.

And I was able to write a 500K LOC app using this methodology and it's maintainable. Can Copilot/Continue/CC do this ? I have tried and they generated crap that's unmaintainable long term. Did you succeed writing a big app with these tools ? I did not but maybe it's possible ?

My app doesn't have downloads any way. So you must be right: It seems useless. Yet it made a difference for me. I don't know...

I built AICode, a VSCode extension and methodology for writing AI-generated code that remains maintainable in large projects.

It started as a manual workflow on ChatGPT.com: assembling prompts in a notepad, iterating on specifications, self-checks before and after coding. Once the system became reliable, I packaged it into a VSCode extension.

The goal is not speed, but producing structured, high-quality, long-term maintainable software using AI. I use it in my own projects, including apps of 500K LOC.

Looking for feedback from devs who care about AI-assisted development and code quality.