This is what people aren’t understanding when they downplay LLM written code. The CAD program I’ve built is the result of 145 Claude sessions to date, and hundreds of hours of testing and feedback
I’ve been working on a budget app for a few months now. It’s every bit as good as what mint.com used to be. But it’s also taken a lot of time and steering to get there. I’m probably 99% done and only making minor tweaks or ui changes at this point.
My next step is a dedicated android app.
I don’t think you realize what you’re dealing with tbh. But I hope you are right.
What do you mean by that?
I don’t understand how the project owner guarantees clean room implementation if they’re accepting PRs? Wondering whether accepting PRs written by legally malicious contributors will legally expose the project, and will result in take down?
Personally forked and locally cloned FilmCraft (403K lines of Rust 
), mainly to learn extreme vibing, but also to try out. The project implements codec encoder/decoders and has web assembly binaries with command based implementation/api optimized for AI automation. Neat stuff. Suspect some Adobe Lawyers and IP lawyers generally are super excited about extreme vibing right now.
Like it could be that I’m Brainwashed by FreeCAD and OCCT because in that community everyone keeps telling me it can’t work as good as parasolid we don’t have the funding or the knowledge we need a team of 100 devs. And everyone says OCCT is the best we can have they spend years in ignoring issues in FreeCAD. I mean they say they spend years in developing the kernel and it would take years to improve it it too parasolid level. Also tpn, FreeCAD spend years promising they where gonna fix it, but nope still sucks.
Please prove everyone wrong…
Yeah I would say that’s all out the window now with AI. It honestly doesn’t make any sense that FreeCAD would still be as bad as everyone says it is at this point. Gotta be either poor management or grumpy anti-AI developers
The chairman stopped being a maintainer because of AI use and instead of forking OCCT and using AI to fix it, they had hired a dev to negotiate with OCCT to fix issues…
The complete FreeCAD team (FPA) are 13 developers. That are probably 10 more then necessary to currently manage it if you’d adopt AI. So there isn’t a big incentive to do this.
FreeCAD is already in damage control, basically trying to hide how much they f’d it up.
Well, at the moment, most open source developers aren’t paid, and AI doesn’t work for free… So how do you intend to bridge that gap?
It’s really easy for people to sit here and say “let’s just vibe code everything” while these companies are still burning cash and subsidizing your use.
But when that runs out, and they have to be profitable, the amount of people that think it’s cool to burn their own cash to vibe code something is going to dwindle.
People won’t spend $20 in total to support massive world-class free software like Blender, but you expect them to burn $20-$50+ per month(and more when the rug gets pulled) to help everyone else?
What you get for $20 today with Claude will be $200+ when they go API rates like Copilot.
I think it’s a fad that will die. When the “oh look how cool it is to create something” part is over and people are left to maintain software, it’s going to get scary. The people who put their life for many years into projects just to have people come along and say “They are terrible because 11 years ago they couldn’t do what I can do today with AI” are going to get those type of people to stop altogether.
I see it as more as a risk of killing open source, rather than enhancing it.
It dilutes the pool too much. Instead of 1 stable frontrunner, you end up with 100 competing things, all with the same bugs, etc.
It’s the reason why I quit trying to do anything with the WebUI. It turns into “I don’t like where this label is on the screen, so I’m just going to vibe code my own UI”. It’s just not worth the effort.
Open Source was always a thankless job… feels even worse nowadays.
About the thankless job I totally understand, and you are right about it’s going to be expensive. But the thing with vibecoding atm is you are not making ai depended software you are making standalone code and standalone software. If anything this could safe us when ai gets expensive as you probably won’t be able to vibe Photoshop anymore without paying. So having independent Photoshop clone is pretty useful.
I said it on the FreeCAD forums too or GitHub that they should abuse the shit out of ai because it is so cheap and that probably wont last. Also setup a independent LLM run by FreeCAD for FreeCAD but the anti ai lobby is so big that they are just stuck into limbo.
But you aren’t making developers…
So you are stuck with large codebases of stuff that sit outside the capabilities of the people who created it.
I have already had actual software developers, people who do this for a living, come to me after they have used AI to create large chunks of code to ask for more AI budget because they can’t understand or support the code AI put out.
The only way to maintain these things properly, from what I’ve seen, is with AI, especially if people are throwing Opus at everything and getting massively convoluted codebases.
I still need to test my local LLM stack against the FreeCAD MCP. Maybe I’ll do that sometime this week.
Yeah. We’ve had a few PRs at work… “Why did you create a new function for this in this file, when there already was one over ‘there’”.
Sort of this thread compressed into a video. He has a lot of points that are talked about here. Also an interesting projection as to where corporate apps like Adobe and such are going to go.
I am wondering if this new AI vibe coding is similar to when we switched from assembly language to higher level languages. There were complaints that the high level languages produced bloated slow assembly code that could not be maintained.
With how fast computers have become an how complex applications have become no one checks assembly code any more. Is it possible code will become irrelevant as well. I have seen talks where they say the spec will be all that is needed and the code will just be regenerated when the spec changes. I am not sure we are at that point yet. Does it really matter if there is duplicate code if the app actually performs to the spec. Just like how the expert assembly language coders dismissed higher level languages as inefficient the expert high level coders so too want to dismiss the AI code generation as a fade. But the AI is getting better and better with every new model drop and as people are using multiple agents and throwing all kinds of computing power at it it only makes it cheaper not more expensive for us hobby makers. I can use a $20/month plan and play all day every day, if I want, on making the tools I want.
Just like high level languages did not kill software development but actually made it more mainstream, I am hopeful that AI coding will not kill software development but make it so even more complex and useful apps can be created, otherwise I as a professional software developer will be out of a job.
It’s not.
Switching to a higher level language still required an engineer with knowledge to create things.
I can almost guarantee those “talks” were from people with a vested interest in selling you either GPUs or tokens…
I don’t see how this correlates to “cheaper”.
Well, I got news for you… it’s coming. Especially if your wish is
Why, as a manager, would I pay someone $100k+ to take a spec and throw AI at it? Why wouldn’t I just get another machine and spend the $100k on tokens?
Developers today who are using AI in lazy ways, are just proving to their bosses why they don’t need them. The more you separate yourself from the implementation, the more you are proving to your bosses why you aren’t needed.
AI can be an amazing tool to enhance value and efficiency, but as soon as the human element isn’t needed anymore, you won’t need high paid developers. You only need maybe technical writers and a QA team.
I think you are missing the point…
" A December 2025 analysis by CodeRabbit of 470 open-source GitHub pull requests found that code that was co-authored by generative AI contained approximately 1.7 times more “major” issues compared to human-written code. The study revealed that AI co-authored code showed elevated rates of logic errors, including incorrect dependencies, flawed control flow, misconfigurations (75% more common), and security vulnerabilities (2.74x higher). Additionally, they also reported high code readability issues, including formatting errors and naming inconsistencies."
" In early 2025, GitClear published the results of a longitudinal analysis of 211 million lines of code changes from 2020 to 2024. They found that the volume of code refactoring dropped from 25% of changed lines in 2021 to under 10% by 2024, code duplication increased approximately four times in volume, copy-pasted code exceeded moved code for the first time in two decades, and code churn (prematurely merged code getting rewritten shortly after merging) nearly doubled."
" In February 2026, GitHub acknowledged an increase in lower quality AI-generated contributions overwhelming open source maintainers, describing the phenomenon as an Eternal September for open source. They cited examples of projects already affected, including the cURL project ending its bug bounty program after AI-generated security reports multiplied, and some projects, such as Ghostty, moving to invitation-only contribution models."
None of this is good for our industry.
It’s terrible for open source.
It commoditizes software development to a point that it will kill it as a valuable, high-paying profession, and overall, means that us users may get stuck with worse, and less supported, options overall.
I just don’t see any scenarios where it’s better for all parties involved.
The biggest disconnect for me, is just how excited everyone is to share vibe-coded things that they don’t have the skills to make themselves.
“Look what I made”… except you didn’t…
It’s just not impressive to me…
As a “creator” and an “engineer”, that’s what I like to do. Make things.
There’s just nothing fun to me about directing AI to do work all day.
If that’s what being a software developer was about 25 years ago, I’d have picked a different career.
Yeah, I probably don’t have the skills for much of what I vibe. But, I definitely don’t have the time to churn out all the unfinished, pre mvp, non-income generating token burning apps I’m currently able to actualize.
I like making things, and appreciate living at a time when that doesn’t require coding via punch cards.
I think it’s great that people can make what they want.
It’s different when it goes public. And worse when it’s attempted to be passed off like it wasn’t 100% AI generated.
I just don’t have high hopes for the path we are headed down
Is it worse if it´s AI generated?