Is it a wrap for Software Engineers? Devin autonomous AI software engineer...

[null]

Read a story once about an AI that diagnosed diseases. This was 10 years ago.

They put the symptoms in

and the AI reads

EVERY medical article published. Ever. In like.. 35 seconds.

Then it tells them the likely disease.

Okay. Great. This is something humans can do. Except the patient had hours to live and the humans would take days to do the research.

The computer was exactly right also.

If the AI can save countless hours of physically typing then that's actually great. Frees the humans up to imagine and create at a faster rate.

 

[null]

I think an unintended side effect of this tech is that people will actually have less expertise.

Like the first tweet says, you need to be a Senior to understand and write the best code with it possible. That includes fixing code it's outputting that you know is wrong.

However, Seniors really found out those shortcomings because they've gone through what every single profession must go through: the learning cycle.

You write code, shyt doesn't work, you got to fix it and understand what went wrong. With that piece of knowledge and lessons, you can now find the error in any piece of code about that topic whether written by a person or A.I. People, especially Juniors who are going to use A.I for like 99% of everything won't have the experience of fixing and making much of anything from pure scratch so they won't understand if what the A.I is doing is truly correct or not. They'll more or less accept it without verifying if it's optimal or even correct.

It'll definitely have impact on writing code no dount, but it'll make worse engineers, codebases and some very costly bugs imo without people who know what they're doing. Unfortunately, I think the longer people use it without keeping their own skills sharp, the lower their ceiling gets.

It really is like the calculator: if someone studies up and knows their material, they can do all kinds of crazy shyt with it. The normal person with weaker math skills can still add do some operations, but nothing advanced.

it won't even compile or worse translate if devs just took what AI generates.

i'm not sure you appreciate the level at which programming instructs computers.

"some" bugs

 

[null]

Shyt is crazy. Claude AI can do the code, you only really need to maintain it.

why? claude can maintain it too, right?

 

[null]

yeah like i said look at the camera in the first iphone compared to a iphone 15

ai is only going to get better and better and better

like self-driving cars are now the norm

 

[#1 pick]

white collar workers bout to feel how unskilled laborers felt at the turn of industrialization

we made machines to make us obsolete in the workspace but they still want us consuming at peak rates to drive these economies. that shyt don't add up...

*disconnect

this shyt won't end well. hope y'all prepared for the chaotic future.

It's already happening

 

[bnew]

Within 40 min codex-cli with GPT-5 high made fully working NES emulator in pure c!

https://archive.is/w257C

Spoiler: text

Posted on Fri Sep 12 23:30:22 2025 UTC

https://i.redd.it/vhuiper0jtof1.png

Within 40 min codex-cli with GPT-5 high made fully working NES emulator in pure c!

Is even loading roms.

I thought AI will be able to write NES emulator not faster than 2026 or 2027 .. that is crazy.

https://i.redd.it/7oz0b3jjjtof1.png

https://i.redd.it/63w5h2jjjtof1.png

GITHUB CODE

GitHub - Healthy-Nebula-3603/gpt5-thinking-proof-of-concept-nes-emulator-: gpt5-thinking proof of concept nes emulator

 

[bnew]

Cursor AI CEO shares GPT 5.2 agents building a 3M+ lines web browser in a week

Spoiler: text

Posted on Sun Jan 18 15:12:16 2026 UTC

Loading…

v.redd.it

Cursor AI CEO Michael Truell shared a clip showing GPT 5.2 powered multi agent systems building a full web browser in about a week.

The run produced over 3 million lines of code including a custom rendering engine and JavaScript VM. The project is experimental and not production ready but demonstrates how far autonomous coding agents can scale when run continuously.

The visualization shows agents coordinating and evolving the codebase in real time.

Source: Michael X

https://twiiit.com/i/status/2012825801381580880 | https://nitter.poast.org/i/status/2012825801381580880 | https://xcancel.com/i/status/2012825801381580880 | https://twstalker.com/i/status/2012825801381580880

 

[Regular Developer]

Cursor AI CEO shares GPT 5.2 agents building a 3M+ lines web browser in a week

Spoiler: text

Posted on Sun Jan 18 15:12:16 2026 UTC

Loading…

v.redd.it

Cursor AI CEO Michael Truell shared a clip showing GPT 5.2 powered multi agent systems building a full web browser in about a week.

The run produced over 3 million lines of code including a custom rendering engine and JavaScript VM. The project is experimental and not production ready but demonstrates how far autonomous coding agents can scale when run continuously.

The visualization shows agents coordinating and evolving the codebase in real time.

Source: Michael X

Michael Truell (@mntruell) | https://nitter.poast.org/i/status/2012825801381580880 | https://xcancel.com/i/status/2012825801381580880 | https://twstalker.com/i/status/2012825801381580880

I saw someone on reddit create the start of an OS using Agentic AI. Not ship ready, but I thought that was cool.

 

[bnew]

I saw someone on reddit create the start of an OS using Agentic AI. Not ship ready, but I thought that was cool.

this tech is gonna be a boon for open source software, privacy minded users and advocates are gonna have even more software options within 5 years.we need larger context windows.

 

[MasterOfAllHeSurveyz]

Mafukkas with the least amout of tech background gas up Ai the most and refuse to listen to common sense,

Realest comment in thread. I use Claude & about half a dozen other LLMs daily. You have to handhold Claude & any other LLM because of hallucinations/limited context windows.

 

[bnew]

Spoiler: thread continued

Spoiler: full text & large images

1/3
@mntruell
We built a browser with GPT-5.2 in Cursor. It ran uninterrupted for one week.

It's 3M+ lines of code across thousands of files. The rendering engine is from-scratch in Rust with HTML parsing, CSS cascade, layout, text shaping, paint, and a custom JS VM.

It *kind of* works! It still has issues and is of course very far from Webkit/Chromium parity, but we were astonished that simple websites render quickly and largely correctly.

[Quoted tweet]
GPT-5.2 Codex is now available in Cursor!

We believe it's the frontier model for long-running tasks.

2/3
@jackccrawford
44,000 commits in one week ... whoa.

3/3
@rohitxdotdev
Bro commited every single line change


To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[EzekelRAGE]

3M lines of code sounds like a nightmare to go through if you have a problem.

 

[MasterOfAllHeSurveyz]

Spoiler: thread continued

Spoiler: full text & large images

1/3
@mntruell
We built a browser with GPT-5.2 in Cursor. It ran uninterrupted for one week.

It's 3M+ lines of code across thousands of files. The rendering engine is from-scratch in Rust with HTML parsing, CSS cascade, layout, text shaping, paint, and a custom JS VM.

It *kind of* works! It still has issues and is of course very far from Webkit/Chromium parity, but we were astonished that simple websites render quickly and largely correctly.

[Quoted tweet]
GPT-5.2 Codex is now available in Cursor!

We believe it's the frontier model for long-running tasks.

2/3
@jackccrawford
44,000 commits in one week ... whoa.

3/3
@rohitxdotdev
Bro commited every single line change


To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

CEO was bullshytting.

=====

In mid-January 2026, Cursor CEO Michael Truell announced the "GPT-5.2 browser experiment," claiming a week-long autonomous run produced a 3-million-line browser in Rust. However, skepticism is high regarding the codebase's actual utility versus its marketing "vibe."
Technical reviews and community teardowns have identified several factors suggesting the project may contain significant "AI slop" or stubs:

• Compilation Failures: Independent analysts and reports from January 16–18, 2026, indicate that large portions of the generated code do not compile and lack basic engineering logic.
• Dependency Reliance: Critics argue the "from-scratch" claim is misleading. Examination of the dependencies revealed heavy reliance on existing open-source engines like Servo and QuickJS for core functionality, rather than entirely original logic.
• Performance Issues: Reports suggest that even when segments run, pages can take up to a minute to load, indicating inefficient architecture or unoptimized "hallucinated" code.
• Non-Functional Components: Some technical deep-dives on platforms like Hacker News and Reddit highlighted that specific complex modules, such as the 1-million-line custom JavaScript VM, are present in the codebase but not actually executed during the rendering process.
• The "Vibe Coding" Critique: The project has been cited as a primary example of "vibe coding," where high volumes of code give a superficial impression of progress without meeting production-grade standards for maintainability or correctness.

While the project demonstrates GPT-5.2's ability to maintain architectural context over thousands of files, current consensus suggests it is more of an autonomous agent benchmark than a functional, "from-scratch" software product.

These technical analyses investigate the Cursor AI browser experiment's compilation failures and reliance on existing libraries:

Cursor AI Browser Experiment: The Vibe Coding Crisis | byteiota

byteiota.com

Software developers don't have to worry about AI taking their job anytime soon. Sam @ OpenAI, and his ilk, are the ones who started this bullshyt con, in the last few years, with AGI. Not happening. We don't even know how our minds work. Now, a person using AI to enhance their skills? Yes, that muthafukka is taking your job. You're the framer using a hammer and nails. The guy using AI has a nail gun.

------

I literally have to constantly remind Claude to obey what I've written in project and user level CLAUDE.md files and use the Claude Mem project to keep a permanent records of our conversations because so much gets lost with compacting when the context window gets full. When it or any other LLM compacts, it makes a summary of what you're trying to accomplish and is pretty good the first run if everything from a single task is completed before compacting again. I and others have pointed out how it'll make a summary of your summary and so much is lost if you don't use skills like plan/tasks/etc. like what's found in an addon project like Spec-Kit.

 

[Mission249]

Most of these "AI creates blablabla from scratch!!! " articles end up being AI either delegating to a dependency doing the heavy lifting, or AI regurgitating some open source code.

I use AI daily successfully, but the hypers and grifters are getting over the top.

 

[cyndaquil]

AI is a very useful assistant for people who actually know what they are doing in the domain they are doing it in. Issue now is we need to rethink what an entry level developer is. A lot of the problems juniors would solve AI is doing it in seconds. But we need those junior level jobs becauss 10 years from now who will be the senior developers?