I do some light coding with Claude, but it’s really not hard to extrapolate to being able to code a fully offline app that makes no calls to anything, or a self-hosted site that functions the same without the trackers, etc., or a frontend for something to kill the tracking aspects. Personally, anything I’ve thought of was already on Fdroid.
Just wondering what others are doing.
This light coding you speak of, are you sure this isn’t vibe coding?
Most apps have already been built for offline usage. I can boot up Linux and do any basic computing task with the installation of some software.
The fact of it is that most useful computing is entirely over the wire, and the problems that need solved are hard ones. Networking and communication is the most important issue in my opinion, and the best we have is Signal, as well have privacy respecting services. But to host online applications for other people cost money, so someone’s gotta pay for that. And if the Tea app has anything to show, it’s that security and privacy still some human expertise and judgement required to make that a reality.
The next fact is that the world is lurching towards utilizing the web in a way that is invasive. More offline apps won’t help if you need to perform computing over the wire.
To continue, self hosting is still a fringe concept. Even working in tech, I’ve yet to me other people who self host on a VPS let alone their own hardware.
I believe the next generation of privacy apps must be paid E2EE centralized alternatives that offer the same UX as existing solutions. That’s pretty hard. Signal is easy enough as a messenger for example. But the technical chops required to not screw that up is also high.
I’m ignoring federated and decentralized apps, as the barrier to create those and keep them secure is even higher, so E2EE centralized solutions are the simplest of the three to create in my opinion.
Eh - it depends on the definition I guess. When does saving time writing HTML and JS turn into vibecoding? Do I need to take and pass a test of my own knowledge to assure others? AI vegans will shout at me that if I even know what Claude is, I’m vibecoding and somehow being disingenuous while also destroying the world.
Thanks for your thoughts - great points.
The use case I was thinking of off hand was some sort of heinous privacy-hating Android app and just pulling the useful features out. Personally, my only need would be a tracker for going for a run that just maps the GPS points every 10 seconds or whatever and does time/km math. I don’t think CoMaps has that option. But it doesn’t need to touch the internet to do any of that, and workout/health apps are notoriously some of the worst at passing data to every company on earth. Just as a preliminary idea (and maybe it’s already in the Fdroid store and I haven’t found it)
I would say if you are capable of building the app/site yourself without AI then it is probably not vibe coding. However, I’m guessing that is not true for most people using AI right now 
You want this: OpenTracks (Reproducible build) | F-Droid - Free and Open Source Android App Repository
Yeah, but where’s the dividing line? It’s entirely arbitrary. For example:
I know Excel somewhat wel’m not where I used to be. Let’s say I get a small data set I need to clean up and then do some formulas, and why not, do a regression on some of the data. It’s been 15 years since the last time I needed to get this deep into Excel. I spend 3 minutes with Claude and have it slap the formulas in. Or I can search for 4 sites that have the formulas I need and do it manually, and it’ll take maybe 30 minutes to get it done.
Have I just “vibecoded” this Excel sheet?
Ahh!!! Thanks so much! I had searched for tons of things, and “sport tracking buddy” was not on my radar to search for.
I always think about how The French Dispatch describes modern art:
One way to tell if a modern artist actually knows what he’s doing is to get him to paint you a horse, or a flower, or a sinking battleship, or something that’s actually supposed to look like the thing that it’s actually supposed to look like. Can he do it? […] The point is, he could paint this [classic portrait] beautifully if he wanted, but he thinks this [modern art] is better.
I think it’s similar here: Basically, prove you’re an Excel wizard first
Proton recently has come under some fire for using Cursor AI, but I’m not particularly concerned about the quality of their code, because they’re Proton and they built Proton Mail for over a decade without AI. If you’re coming right out of the gate with your Cursor-generated product on the other hand, then you’re vibe coding lol
I agree with this assessment 100%. Where I think the lack of alignment lies is simply the nuts-and-bolts practicality of it all vs. the stigma and internet outrage.
Case in point, the Cursor use at Proton: It’s entirely possible that this was as simple as 1 intern using Cursor and their supervisor not doing a thorough review of their code. Or Proton doing a review of Cursor to see if the results could work for them to save money and time. And while Cursor is obviously an AI-first IDE, it’s entirely possible to use Cursor without an LLM. But the stigma with lack of understanding is what churned speculative outrage and pitchforks to come out, as if this endemic to Proton and the whole company was magically now a 22-year-old who “made an app” that was spaghetti code and exposed databases.
None of this prevents anyone from “vibecoding” or using LLMs for one of the 8 things they’re actually good for. What it does is encourage obfuscation, or force tribalism so that anyone willing to say how they got the code can only do so within a large community of other people who have the same expectations.