I think it comes down to where you want to spend your time and attention.
Like, I have to imagine that 80% of the "quick stuff" people need to do in Blender would be done with:
* create quad * exude like 5 times * done
Maaaybe
* perform boolean operation between object and a circle or 3
---
Like, I have less than 200 hours in Blender and I was pretty functional after like ... 30 minutes to an hour of a single tutorial.
You’ve likely just switched hobbies to your Blender clone instead of what you wanted to make.
A good visualizer might be useful thought! If that part of the work needs to be validated.
I wonder if Claude Code and other similar tools start to maintain their own base apps that everyone keeps asking for? This would be like a "cache" that makes the requests cheaper/easier to deliver?
Then everyone's "own" apps is really just a customization layer on top of a solid base that already does all the heavy lifting.
All these clones of Photoshop and similar tools are mostly dead ends, they are experiments to see if you can do it, but there will be a consolidation around regularly maintained and powerful and complete apps that likely also offer support.
These experiments are dead-ends for the most part long-term. Fun play things, but they will not gain significantly user bases.
Why? Every time someone cripples software in the name of safety it just gets replaced with more capable software that isn't.
For the rest, at least so far, I haven't had the collaboration experience but for word processors I can't go back to local files that can't easily be edited/viewed/commented on by multiple people
I think they all have been post-trained on specific tasks like building sites, apps, etc and that is a large part of the reason they are pretty good at it.
I think the value of adapting full tools to your own workflow far outweighs the maintenance costs nowadays. The tool above for instance took my agent a couple hours and it was good to go. It wasn’t a small slice by any means either: it supports edit/obj mode, extrude, loop cuts, modifiers like subdiv, save/load, undo/redo, bones, auto rig, posing, and more.
Mind you, this is a slice of an open source tool which was still outside of what I wanted from it. Cost wasn’t even part of the equation here.
Since Blender is OSS, you could probably use one of the existing VR views for Blender and write a bunch of in-app Python scripts to do what you want, if you wanted to also be able to output to Blender files and work within Blender already.
This looks like a pretty well-trodden area in Blender.
Adding scripts into an existing product does not serve this.
I think we are entering into a new era where we don’t need to accept fully featured software with all of its decisions should they not fit our wants
And you get battle-tested software and a huge collection of existing plugins, operational modes, export formats, all that. Fully bespoke software for everything you want is a way toward losing interoperability between people and platforms.
But building up as opposed to building down is freeing in that I can also pull in ideas from Houdini and mix them in. Not to mention BECAUSE Blender is OSS, i can also use it as reference for implementations of particular things should I need that too.
IMO, this kind of full customization is going to become far more commonplace.
One of the tools (built for me to score options of different assets): https://id-le.com/review/intro-logo-animation
Video: https://youtube.com/shorts/tlI9_XBuqVk
Repo: https://github.com/nicobrenner/id-le
Online: https://id-le.com/
I could start from scratch obviously - but seems a shame.
I think both those apps are great and happy to know folks like you are maintaining them; this one though is simply more of a personal software than a path to an OSS project!
Before that, though, I think you have to learn what the experts do and what can be done and what you should want. And then you can start to make your own opinions.
By then, unless the major players have done it really wrong, you're not going to make something better than what already exists.
Right now, it's different, because we have that experience prior to the ability to just wish up an app and have it happen.
It’s always more work than it looks. You can get to mostly usable tools quickly if your request isn’t too difficult, but every time I use them I run into another thing that needs to be fixed, changed, clarified, or cleaned up.
It’s fun at first, but over time you realize you’re spending more time building the tools than using the tools to build the thing you originally wanted to work on.
I think personal software is best for highly specific tools or for things you need to use repeatedly for a very long time.
For something like a SketchUp competitor, I wouldn’t want to try to maintain my own. The number of edge cases and features is bigger than you imagine up front, and the vibecoded version that might appear a few hours after your prompt is going to be very far from the version you want to use.
My personal projects matter a lot. For some of them I am trusting my private data in the code, so I need to hold it to a high security and privacy standard. For others, I have hundreds or thousands of users (game mods, helper tools for hobbies, etc.), and I don't want to ship low quality to them.
Because I offer no warranty and work on it on my personal time. What use is code nobody uses?
> Also why would your private data leak just because you used AI to reduce time to code?
Because if I don't exercise rigor and review oops you let the security holes in (I had to do like 4-7 separate pentest passes on AI generated code to catch and close the vulnerabilities)
As I said, that is not what I meant by "personal projects that don't matter". And who said about code that nobody uses? 1 person is using that code in my definition, in case it wasn't clear.
So we are just talking semantics here at this point.
If it doesn’t matter then do what you want. By definition, it doesn’t matter.
Some of us actually use these tools, so it does matter. You do have to consider things when it matters.
The second sentence I agree, of course you need to consider things when it matters. Thats the basic definition of importance.
AI can get these done implementation wise but in my experience it needs guidance because it can't guess what features and details I want /don't want.
The difference is that it's now possible at all for me but still requires a few hours every now and then supervising.
It’s amazing that you can get something usable quickly, but then every time you use the software you’re back in the AI working on your tool.
Using AI to build the tool becomes the project, replacing the project that led you to think about vibecoding a tool.
For instance in the context of this SketchUp clone that might mean using AI to generate an addon for Blender.
For some of my recent projects I'm using the community edition of Odoo to serve as that substrate because it already has an established community of modules to draw from, ten years or so of real world use, and I don't have to burn tokens generating a user account system with permissions or the data models / ORM to interface with them.
This seems to be serving me well, but we'll see. It could be that I end up in the same the tool becoming the project scenario that you describe, just a little bit later than if I had started from a blank slate.
The vibecoded apps all rely on an army of users exploring the program and reporting all the bugs and edge cases. You don’t want to be doing that on your own.
it seems like the open-source space is losing some aspects of collaboration. people used to search for already existing projects on some git-hosters, maybe fork them or put effort into the existing thing. now half a dozen people each run their own race using AI, each solution misses 30% of the features and has x-amounts of bugs.
work together ffs :( i mean, 20 people can use AI to vibecode the same project, why does everyone now feel the need to do everything solo. (because they want to be the prolific founder and cool dude who did the most in the project* instead of being one of many contributors, i assume...)
*or rather, their AI.
I'm not too sure what the future here looks like and if this is even a problem, but things certainly are changing. These days submitting a well written bug report or feature suggestion, or validating existing reports is more useful than sending a PR.
ArtCraft is definitely a play at selling subscriptions.
The slew of "oh yeah, blah blah blah blah blah, that's why I built [X]." comments on every AI post now?
Click through and many you'll find "Pricing".
It's a gold rush.
2. Click "Fork"
3. Profit
Right, and that's something I already argued for since the very beginning of GenAI, namely that suddenly being very costly, economically and ecologically, is not that bad if you build upon what exists and collaborate on progress.
... except that's exactly WHY vibecoded project exist : they do NOT have to give a shit with existing codebase and agree on weird rules. They just start from "scratch". In fact I'm convinced there is the Ikea effect at play, namely that the PM of the vibe coded project is emotionally attached, even proud, of the result they think they provided somehow, despite them not actually doing it. Obviously for-profit GenAI companies have a big incentive to maximize token use and will never promote workflows and tool that minimize the one thing they monetize.
So even though I agree in theory, in practice I think there is little to no overlap.
To be fair neither is quite as complex as a whole 3D modeller.
> To be fair neither is quite as complex as a whole 3D modeller.
Once you cross a certain complexity in your project, it becomes quite work intensive even using AI to do the coding. All the back and forth and cleaning up, exploring ideas and starting from scratch again takes time and effort and maybe most importantly, extensive human thinking. Just to clarify, I am not against AI and use it extensively myself. But vieb-coding all your tools is just not really a feasable idea for most (time/token-cost constraint) if you tools are of a certain complexity.
But you could easily vibe code, e.g. an H.265 decoder, even though it is super complex.
I have lot of tools that work for me.
When externalities break them, it's usually one prompt to fix them.
https://m.youtube.com/playlist?list=PL3NwjxPeybkLhFjSfFg1iL4...
This is precisely what I'm doing.
Would love to hear advice from people who use CRM tools on the job.
By outsourcing the modeling to Claude or whatever, you don't learn how to build 3d models on your own. You've become dependent on the LLMs, and you're trivially replaceable by someone who can build 3d models on their own...because they can also learn to tell Claude to do all the work.
Their models are horribly topoligized, which will matter for most use cases for models; and if you’re just using it to make a cube, it’d be better to learn the related program to do that.
People tend to ask LLMs to make full art pieces (a decade of specialized knowledge and muscle memory), not “draw me a square”.
In 3d art, there is a gulf between the easy stuff and the hard stuff.
Tl; dr: for 3d art, I agree.
Not necessarily. But only if what you need is not available in open source or not achievable by forking open source. Otherwise it’s a waste of money (token) and attention. This is a significantly smaller pool that “build [all] your own tools”.
Although, AI mega corpos would love if we do constantly re-build everything. You can tell by the tendency of the agents to write everything from scratch instead of importing the code that does the same thing or better.
At least that's my experience.
Small tools with few features is great. a 3D modeler is not that.
It gets so much worse. At some point you start hitting edge cases where there is no clear right answer and you have to come up with a kernel wide coherent approach to things like 1ulp differences. Parasolid is full of decades of kludges on top of kludges and giant switch statements trying to preserve backwards compatibility as they explored the space.
I suspect people saying "everyone will vibe code all their software" have never actually tried to vibe code a meaningfully complex application.
Wat is the point of personal projects if in the end you learned nothing
https://news.ycombinator.com/item?id=50406927
For an example. Lots of people are now vibe coding their tools.
I think people underestimate what a time investment building your own tool that's of at least medium complexity is, though. It might be worth it if it's something you'll use regularly for a long time, but I don't foresee it being the norm. Even assuming no bugs ever or no performance issues ever and all the architectural decisions made were appropriate (pretty far from current reality), merely describing what you want and how it should behave and look like is a long process, and one that isn't all that enjoyable once you go through it once or twice already.
The existing software that has been around for several years and you kinda dislike is probably much more reliable than your own homebrew, too. Doesn't matter in some cases, matters hugely in others. Learning that your custom video gallery organizer just vanished half your family videos probably isn't very enjoyable, but at least it's easy to recognize it happening. Something more abstract like CAD or 3d printing slicer, ehhh...
I have no doubt as a newcomer you can make software that will do what you want, but most likely the outcome will no be good, not because of yourself or the tool, but rather because the expertise to map the problem space to the solution space is not trivial, and that's precisely what requires a new way to think, and thus tools that aren't so easy to learn. It's not poor UX.
"So many times I’ve had to spend hours learning a software, reading docs, doing tutorials, only to later be able to do a small basic idea of what I actually wanted to do in the first place
Now I can just build only the software I need, for the thing I need it for"
So, I guess yes, if there is someone's hard work you can copy and steal and make your own, I guess that's true. But try to create something that no one has ever built yet.
I host it here: https://vibe3d.spoerri.dev - my partner enjoys it because she can bring an idea to print pretty fast.
Maybe a similar idea could be interesting for KetchUp as well.
Need a 3D modeler? Use Claude or codex to build the 3D model first, then build the modeler around it as you iterate towards what you want to build
So many times I’ve had to spend hours learning a software, reading docs, doing tutorials, only to later be able to do a small basic idea of what I actually wanted to do in the first place
Now I can just build only the software I need, for the thing I need it for (or, ala REA, maybe either modify or reverse engineer some existing software, to better fit my workflows)
In that sense, it would also be cool to be able to drive KetchUp with an agent. I’m a big fan of what pen.dev has done (although now their tool is not free/open anymore)