You can get even more vague and just generally describe the design of something, making sure it leaves exact measurements to parameters, and end up with something usable. ("Make me an openSCAD file for an pointed star with curved points and an inward taper. The number of points, thickness, and angle of taper should be configurable")
Solved with direnv. Also - in my .bashrc in all of my (many) clients:
$ type uvi uvl uvv
uvi is a function
uvi ()
{
uv pip install $@
}
uvl is a function
uvl ()
{
uv pip list
}
uvv is a function
uvv ()
{
uv venv;
cat > .envrc <<EOF
source .venv/bin/activate
EOF
direnv allow
}
Because the programming is and was always a means to an end. Obsessing over the specific mechanical act of programming is taking the forest for the trees.
I agree with gp that the speed in which I am able to execute my vision is exhilarating. It is making me love programming again. My side projects, which have been hanging on the wall for years, are actually getting done. And quickly!
The actual act of keying in code is drudgery for me. I've written so much code in so many languages that it is hard not to hate them all. Why the fuck is it a hash in ruby but a dict in python? How the hell do I get the current unixtime in this language again?!? Why the fuck do I need to learn yet another stupid vocabulary for what is essentially databinding? Who cares, let the AI handle it
None of my side projects are things where I want the output. They're all things where I want to write the code myself so I understand it better. AI is antithetical to this.
I have three side projects that revolve around taking public access data from shitty, barely usable local government websites, and then using that data to build more intuitive and useful UIs around them. They're portfolio pieces, but also a public service. I already know how to build all of these systems manually, but I have better things to do. So, hell yeah I'm just going to prompt my way to output. If the code works, I don't care how it was written, and neither do the members of my community who use my free sites.
The code is still clean and organized. When I say I don't care "how" the code is written, I mean I don't care how it was generated (i.e., by prompting or by hand -- and I don't do any of it manually anymore). I still care about how it's structured, as a well-structured codebase helps the agents do their job better.
Perhaps part of their portfolio is the code they've hand written, and part of it is demonstrating they are able to use this new tool to make something that works (despite how imperfect the tool is, as we see so many people point out)
All of my side projects scratch an itch, so I do want the output. There are not enough hours in the day for me to make all the things I want to make. Code is just the vessel, and one I am happy to outsource if I can maintain a high standard of work. It's a blessing to finally find a workflow that makes me feel like I have a shot at building most of the things I want to.
I wanted a stackable desk tray shelf thing for my desk in literally any size for my clutter. Too lazy to go shopping for one, and couldn't find one on any of the maker sites, so I had claude write me an openSCAD file over lunch break then we iterated on it after-hours. By end of work next day I had three of them sitting on my desk after about 3 hours of back-and-forth the night before (along with about half a dozen tiny prototypes), and thats including the 2hr print time for each shelf.
I want a music metadata tool that is essentially TheGodfather but brought into the modern day and incorporates workflows I wish I had for my DJing and music production. And not some stupid web app, a proper desktop app with a proper windowing toolkit. I'd estimate it would take me 12-18 months to get to a beta the old way, to the exclusion of most of my other hobbies and projects, instead first Gemini then Claude and I managed to get a pretty nice alpha out in a few months over the summer while I was unemployed. There's still a lot left I want to add but it already replaced several apps in my music intake workflow. I've had a number of successful DJ gigs making use of the music that I run through this app. Funny enough the skills I learned on that project landed me a pretty great gig that lets me do essentially the same thing, at the same pace, for more pay than I've ever made in my SWE career to-date.
A bunch of features for my website, a hand-coded Rails app I wrote a few years ago, went from my TODO pile to deployment in just a couple of hours. Not to mention it handled upgrading Ruby and Rails and ported the whole deployment to docker in an afternoon, which made it easy to migrate to a $3 VPS fronted by cloudflare.
I have a ton of ideas for games and multimedia type apps that I would never be able to work on at an acceptable pace and also earn the living that lets me afford these tools in the first place. Most of those ideas are unlike any game I've ever seen or played. I'm not yet ready to start on these yet but when/if I do I expect development to proceed at a comfortably brisk pace. The possibilities for Claude + Unreal + the years and years of free assets I've collected from Epic's Unreal store are exciting! And I haven't even gotten into having AI generate game assets.
Would you share the music app? Do you have a public repo or demo somewhere?
You didn't really describe it very much, so it's hard to say what it actually does. I'm interested in evaluating the quality of vibecoded projects people actually use.
At a later date, perhaps. I haven't messed with this project since I got employed and it was written over summer 2025, when the tooling for agentic development was a lot worse. (Very ADD here) There's also the open question of how best to package a python app that makes use of PyTorch and SciPy for distribution to nontechnical users. I want to solve that before I start putting this in other people's hands.
Careful with the term 'vibe coded', that does not characterize how I work.
Vibecoding is the term for building software with LLM tools. Did you do something different?
I'm just getting tired of hearing claims of incredible software being built with LLM-based tools, but when I ask to see them, I get nothing.
Your claim of 12-18 months for a windowed music metadata app seem weird. That seems like about a week with Dear ImGui and some file format reading libraries to me. Am I missing something?
> Vibecoding is the term for building software with LLM tools
without manual review and guidance. Coasting along purely on vibes. Hence the name. Agentic development is the middle ground where you're actively reviewing and architecting.
Dear Imgui isn't a 'proper' windowing toolkit. It's immediate-mode, it doesn't use OS affordances. Its not WinForms or GTK or QT (though to be fair QT isn't quite native but its by far the closest)
I never made any claims of 'incredible software'. I am building things that I need and want. I will give them to the world if I so choose and if they are good enough. And its not there yet.
And considering that I have almost zero domain knowledge in the area of DSP or audio analysis, that I'd only have a couple hours a day to work on it at best (energy, motivation, and other factors notwithstanding), and the amount of learning it would take to get to the point where something like that would be "about a week" is where most of that 12-18 months goes. And yes the metadata and GUI parts are easy, but the code that generates the metadata that is good enough to perform with? Across every possible container/meta/audio format? That produces quality results on both beatport downloads and 96khz vinyl rips? I'm trying to build something to consolidate my original music library (hundreds of thousands of files) with divergent sublibraries on multiple (proprietary) DJ platforms. Basically cleaning up after 20 years of fucking around without a plan. That's hard.
> The actual act of keying in code is drudgery for me. I've written so much code in so many languages that it is hard not to hate them all. Why the fuck is it a hash in ruby but a dict in python? How the hell do I get the current unixtime in this language again?!? Why the fuck do I need to learn yet another stupid vocabulary for what is essentially databinding?
These are the downsides, but there are also upsides like in human languages: “wow I can express this complex idea with just these three words? I never though about that!”. Try a new programming paradigm and that opens your mind and changes your way of programming in _any_ language forever.
You are doing something, but 99% of the work has been done for you. I guess it's like vibe coding and telling the model to fix issues when you see them.
"I thought using loops was cheating, so I programmed my own using samples. I thought using samples was cheating, so I recorded real drums. I thought that was cheating, so I learned to play. I thought using purchased drums was cheating, so I made my own. I thought using pre-made skins was cheating, so I killed a goat and skinned it. I thought that was cheating too, so I raised my own goats from birth. I haven't made any music lately, what with the goat farming and all."
Yeah, yeah we get it. Does anyone have a different response? Or, equally acceptable, can anyone unpack this one and its siblings?
As far as I can tell, my implicit argument doesn't really rest on some kind fallacious threshold on really doing the thing, or not, right? I never said anything about "cheating." I'm just asking why we got a lot chefs who don't wanna cook.
Like would yall be happy if we maybe just come up with a different word? Why do you need to be an AI user and also "programmer"? What about: "no I don't like coding, but I do love generating apps.."?
Like following the (presumed) logic here we should say that all visual artists are just wannabe photographers. Or we'd might discover an abandoned house in the woods and then feel legitimized to say we built it, because really, the end product is the same right?
AI guys, I implore you, the world is literally yours. Mountains will be moved for your cause. You are the winner, the hegemonic force for the forseeable future. Why can't yall just be happy being what you are? Why cling to titles from an obsolete world? Where you really want to be "programmer" and "artist"? These things have nothing for you, if you believe in this radically different future so much, why even be concerned as being the same kinds of things as us dumb luddites?
I think you should just use your agential blahblah to start a one-person B2B SaaS, make a lot of money, and move on. At no point will it matter what you called yourself!
> Because the programming is and was always a means to an end.
No. Programming is a specific act (writing code), and that act is also a means to an end. But getting to the goal does not mean you did programming. Saying "I'm good at programming" when you are just using LLMs to generate code for you is like saying "I'm good at driving" when you only ever take an Uber and don't ever drive yourself. It's complete nonsense. If you aren't programming (as the OP clearly said he isn't), then you can't be good at programming because you aren't doing it.
I guess I agree with you, but I think the GP may have mispoke and meant he loves building software. It's sort of like the difference between knitting and making clothes. The GP likely loves making clothes on an abstract basis and realized that he won't have to knit anymore to do so. And he really never liked knitting in the first place, as it was just a means to an end.
It’s similar to the arrival of mechanized looms in the 19th century. My ancestors were weavers, and automation eventually replaced those jobs. I’ve spent 40 years working in IT as a programmer and am now nearing retirement, so I’ve been fortunate. To me it feels like programming as a skill may not have much time left. Probably how my ancestors felt.
Yeah, I was reading a little bit about knitting before my post and saw that in 1589, a person who invented a sort of prototype to the automated knitting machine in the UK had his patent application denied by the queen due to taking jobs away from hand knitters. I guess back then they had to be a little more protective because it would be a lot easier for civil unrest to lead to revolution and civil war in postfeudal UK than now.
Most people who are knitting do it purely for the experience of knitting. If you need clothes it's far more affordable to buy the cheap manufactured stuff. Some people certainly enjoy the creativity of expression and wish they could get to that easier - but most of those people have moved away from manual tasks like knitting and instead just draw or render their imagination. There's genuine value in making things by hand as the process allows us time to study our goal and shape our technique mid-approach. GP may legitimately like knitting more than making clothes.
I think you misunderstood my post. Now many people do knitting for the joy of knitting, but people used to knit to create clothing to wear or to sell. Of course, automated knitting machines have largely replaced hand knitting, and people now still do it. If you are very good at hand knitting, you might see if you can sell some work. However, if you want to make knitted clothing at scale, you would be better served taking a high-level approach to the actual design of the clothing and learning how to prompt the automated knitting machine to do so instead of optimizing for how you yourself would hand knit it.
That would be a maximally economically efficient approach to producing knit clothes - but hand knit clothing still does have a significant market. This year I sought out a cobbler to get a new pair of shoes because my feet are a bit weird and the machine templates for what a foot should look like doesn't produce something I can comfortably wear. If you personally derive value from putting in the manual labor to produce "artisanal" goods in most fields you can find a market willing to pay the premium for your labor. This market is much smaller than the machine-driven equivalent so it can't support nearly the same quantity of producers as the market supported before automation came along but it is a niche you can operate within.
I don't disagree with your main thesis that an automated knitting machine can out produce hand-knit goods but I do think you're under appreciating that there still is a market for the non-automated goods. Even if they can't compete for the majority of the market markets are weird and non-uniform so those skills do still feed into a market.
I'm still reading the code, I'm still correcting the LLM's laughably awful architecture and abstractions, and I'm still spending large chunks of time in the design and planning phase with the LLM. The only thing it does is write the code.
But that's not programming because its a natural-language conversation?
> But that's not programming because its a natural-language conversation?
Correct. Programming is writing code. You are not writing code, therefore you are not programming. I don't understand what's so complicated about this.
I'm literally making a program. Present-progressive of the verb to program. I feel like you're pearl-clutching on semantics. By my read, programming != writing code, but writing code is most definitely programming. Oxford defines 'to program' as both.
You're not making a program. You're managing the AI that is making a program. You're a manager, maybe a designer or architect too, but not a programmer.
These are well defined roles that existed well before AI. You don't get to redefine them just because you feel like you should get to be part of some imaginary "programmers' club" without doing the actual thing that defines the "programmer" role.
If you micromanage the mechanic, then yeah you might get production credits for fixing the car.
You could argue that I'm playing the manager, sure. I guess people who write software with nocode or visual data flow tools aren't programming in some form either? They aren't 'visual programming'? What about if I draw buttons and text boxes on a form in Visual Basic? I haven't hooked up the events yet, but that isn't programming?
Would you say that I am not programming if I make a synthesizer in Reaktor or Max? What about using blueprints in Unreal? Are those not programming?
This assertion that programming requires writing code is incorrect. I suspect the distinction cuts a little too close to home, which is why we are arguing semantics here.
It's sad to watch the mental gymnastics at play. I guess by asking my mechanic to service my car, I'm a mechanic too? I want it > it gets done > I am the doer. Ridiculous.
No sane person would argue Person A is a concreter if Person A is telling Person B to do concreting and Person B does the concreting. Doesn't matter if Person A had elaborate plans for the concrete, or if Person A owns the concrete afterwards. These are long-established ideas. You can twist it and argue "semantics!" all you want but it will never take with anyone but the Person A's.
I mean, yes - you’re reviewing and architecting, but not creating.
Same as if you use an image diffusion model. You can describe very clearly what you want, and iterate carefully until you get a picture that looks good. But nobody would say that they “drew a nice picture”, since they haven’t done any drawing.
(except maybe the mega-power-users who use the tool and have a warped view of their accomplishment)
Sounds like you just don't like programming. And that's okay! It's okay to not like things.
But "I love programming now that I don't do any programming" is an utterly nonsensical statement. Please stop and reflect over what you said for a moment.
Maybe, but the amen break has a very specific je ne sais quoi that makes it way more useful and pleasant as a sample than almost any other sample. There's just so many situations in the kind of music I make where the amen is like the only loop that fits. Funky drummer might come in second.
It could just be its cultural weight has me hypnotized. But maybe its just that good
I’ve produced music through much of 2010-2020, I wasn’t there in the 1980-2010s but it wasn’t uncommon see discussion online about different samples or things like this. Never really seen any mention something like this unquantified “je ne sais quoi” or at least don’t really recall
My take is, it was the first of its kind to widely circulate exhibiting desirable quantities for sampling, a combination of good enough and path dependency. After a certain level of saturation/entrenchment it carried an aesthetic compared to readily available samples (maybe this is what you meant).
Whenever I couldn’t find a breakbeat sample (or wanted some starting point at least) I’d default to it. When I did music production it was very easy to get your hands on a loop but obviously that’s much later.
With google covering only 3% I wonder how much people still care and if they should. Funny: I own and know sites that are by far the best resource on the topic but shouldn't have so many links google says. It's like I ask you for a page about cuban chains then you say you don't have it because they had to many links. Or your greengrocer suddenly doesn't have apples because his supplier now offers more than 5 different kinds so he will never buy there again.
> because my prompts are in natural languages, and hence ambiguous.
Legalese developed specifically because natural language was too ambiguous. A similar level of specificity for prompting works wonders
One of the issues with specifying directions to the computer with code is that you are very narrowly describing how something can be done. But sometimes I don't always know the best 'how', I just know what I know. With natural language prompting the AI can tap into its training knowledge and come up with better ways of doing things. It still needs lots of steering (usually) but a lot of times you can end up with a superior result.
Yes. LLMs are search engines into the (latent) space or source code. Stuff you put into the context window is the "query". I've had some good results by minimizing the conversational aspect, and thinking in terms of shaping the context: asking the LLM to analyze relevant files, nor because I want the analysis, but because I want a good reading in the context. LLMs will work hard to stay in that "landscape", even with vague prompts. Often better than with weirdly specific or conflicting instructions.
But search engines are not a good interface when you already know what you want and need to specify it exactly.
See for example the new Windows start menu compared to the old-school run dialog – if I directly run "notepad", then I get always Notepad; but if I search for "notepad" then, after quite a bit of chugging and loading and layout shifting, I might get Notepad or I might get something from Bing or something entirely different at different times.
reply