alt Hacker NewsSo where are all the AI apps?
Comments
My take is you are missing out on a barrage of "Shadow AI" and bespoke LoB and B2B software (By "Shadow AI" I mean the (unsanctioned) use of GenAI in Shadow IT, traditionally dominated by Excel and VBA).
All of the above are huge software markets outside of the typical Silicon Valley bubble.
I’ll ask another question. Why isn’t software getting better? Seems like software is buggier than ever. Can’t we just have an LLM running in a loop fixing bugs? Apparently not. Is this the future? Just getting drowned in garbage software faster and faster?
I think this is a great question to ask and maybe I need my own blog to post about these things as I might reply with a big comment
Making Unpublished Software for Themselves
One issue is, I think maybe a lot of people are making software for themselves and not publishing it - at least I find myself doing this a lot. So there's still "more software produced than before", but it's unpublished
LOC a Good Measure?
Another question is like Lines of Code, about if we best measure AI productivity by new packages that exist. AI might make certain packages obsolete and there may be higher quality, but less, contributions made to existing packages as a result. So actually less packages might mean more productivity (although, generally we seem to think it's the opposite, conventionally speaking)
Optimizing The Unnoticeable
Another issue that comes up is maybe AI optimizes unnoticeable things: AI may indeed make certain things go 100x faster or better. But say a website goes from loading in 1 second to 1/100th of a second... it's a real 100x gain, but in practice doesn't seem to be experienced as a whole lot bigger of a gain. It doesn't translate in to more tangible goods being produced. People might just load 100 pages in the same amount of time, which eats up the 100x gain anyway (!).
Bottleneck of Imagination
I think also this exposes a bottleneck of imagination: what do we want people to be building with AI? People may not be building things, because we need more creative people to be dreaming up things to build. AI is only fed existing creative solutions and, while it does seem to mix that together to generate new ideas, still the people reading the outputs are only so creative. I've thought standard projects would be 1) creating open source alternatives to existing proprietary software, 2) writing new software for old hardware (like "jailbreaking" but doesn't have to be?) to make it run new software so that it can be used for something other than be e-waste. 3) Reverse engineering a bunch of designs so you can implement some new design on them, where open source code doesn't exist and we don't know how they function (maybe kind of like #1). So like there is maybe a need for a very "low tech" creation of spaces where people are just regularly swapping ideas on building things they can only build themselves so much, to either get the attention of more capable individuals or to build up teams.
Time Lag to Adapt
Also, people may still be getting adjusted to using AI stuff. One other post detailed that the majority of the planet does not use AI, and an even smaller subset pays for subscriptions. So there's still a big lag in society of adoption, and of adopters knowing how to use the tools. So I think people might really experience optimizing something at 100x, but they may not know how to leverage that to publish it to optimize things for everyone else at 100x amount, yet.
Social Media Breakdown?
Another problem is, I have made stuff I'd like to share but... social media is already over-run with over-regulation and bots. So where do I publish new things? Even on HN, there was that post about how negative the posters can be, who have said very critical things about projects that ended up being very successful. So I wonder if this also fuels people just quietly creating more stuff for their own needs.
Has GDP Gone Up or Time Been Saved?
Do other measures of productivity exist? GDP appears to have probably only gone up a bit. But again, could people be having gains that don't translate to GDP gains? People do seem to post about saving time with AI but... the malicious thing about technology is that, when people save 10 hours from one tool, they usually just end up spending that working on something else. So unless we're careful, technology for some people doesn't save them much time at all (in fact, a few people have posted about being addicted to AI and working even more with it than before AI!).
Are There Only So Many "10x Programmers"?
Another issue is, maybe there are only a minority of people who get "10x" gains from AI; at the same time, "lesser" devs (like juniors?) have apparently been displaced by AI with some layoffs and hiring freezes.
Conclusion
I guess we are trying to account for real gains and "100x experiences" people have, with a seeming lack of tangible output. I don't think these things are necessarily at odds with each other for some of the aforementioned reasons written above. I imagine maybe in 5 years we'll see more clearly if there is some noticeable impact or not, and... not to be a doomer / pessimist, but we may have some very negative experience from AI development that seems to negate the gains that we'll have to account for, too.
Among the various ways this analysis is flawed,
two that are drawn from my own experience are:
- meaningful software still takes meaningful time to develop
- not all software is packaged for everyone
I've seen a lot of examples shared of software becoming narrow-cast, and/or ephemeral.
That that doesn't show up in library production or even app store submissions is not interesting.
I'm working on a large project that I could never have undertaken prior to contemporary assistance. I anticipate it will be months before I have something "shippable." But that's because it's a large project, not a one shot.
I was musing that this weekend: when do we see the first crop of serious and novel projects shipping, which could not have been done before (at least, by individual devs)... but which still took serious time.
Could be a while yet.
My experience with AI-driven and AI-assisted development so far is that it has actually enhanced my workflow despite how much I dislike it.
With a caveat.
If you were to compare my workflow to a decade ago, you wouldn’t see much difference other than my natural skill growth.
The rub is that the tools, communities and services I learned to rely on over my career as a developer have been slowly getting worse and worse, and I have found that I can leverage AI tools to make up for where those resources now fall short.
The thing to shill now is agents.
So they are all producing products to produce products. My guess is 50% of token usage globally is to produce mediocre articles on "how I use Claude code to tell HN how I use Claude code".
I am still waiting for them.
Cool data. What do I do with it? None of my use cases involve writing software, so I don't think this is _for_ me since my extensive AI use wouldn't show up in git commits, but I'm not sure who it's for. When I'm talking to artist friends, musician friends, academic friends, etc data is nice to have but I'm talking in stories: the real thing I did and how it made me better at the thing.
AI is unbelievably useful and will continue to make an impact but a few things:
- The 80/20 rule still applies. We’ve optimized the 20% of time part (a lot!) but all the hype is only including the 80% of work part. It looks amazing and is, but you can’t escape the reality of ~80% of the time is still needed on non-trivial projects.
- Breathless AI CEO hype because they need money. This stuff costs a lot. This has passed on to run of the mill CEOs that want to feel ahead of things and smart.
- You should be shipping faster in many cases. Lots of hype but there is real value especially in automating lots of communication and organization tasks.
I feel that your assumption that everyone will want to share is a flawed one.
I agree with the premise of the article, in the sense that there has not been, and I don't think there will be, a 100x increase in "productivity".
However, PyPi is not really the best way to measure this as the amount of people who take time to wrap their code into a proper package, register into PyPi, push a package, etc... is quite low. Very narrow sampling window.
I do think AI will directly fuel the creation of a lot of personal apps that will not be published anywhere. AI lower the barrier of entry, as we all know, so now regular folks with a bit of technical knowledge can just build the app they want tailored to their needs. I think we´ll see a lot of that.
I, for one, am not publishing my “apps” for others to use because my “apps” make me money
I am now scared to talk to anyone. Eventually the conversation turns to AI and they want to talk or show their vibecoded app.
I am just tired boss. I am not going to look at your app.
On my local computer used only by me because now I don't need a corporation to make them for me. In the past decades I'd make maybe one or two full blown applications for myself per 10 years. In the past year "I" (read: a corporate AI and I) have made dozens to scratch many itches I've had for a very long time.
It's a great change for a human person. I'm not pretending I'm making something other people would buy nor do I want to. That's the point.
Quite a few AI apps (more like 90% AI apps, surprisingly difficult to get an AI to do anything more than that) are helping educate my kids.
On the one hand, I couldn't hope to do anything close to what I'm doing without AI, on the other hand "write an app to teach me to pass high school exams" is utterly out of reach of current frontier models ...
Like others have mentioned, I think the premise of looking at the most popular few projects (pypi.org currently lists 771,120 projects) on pypi as any sort of proxy for AI coding is terribly misguided/unrepresentative and that almost no one is going to be packaging up their vibe-coded projects for distribution on pypi.
That being said, I've personally put 3 up recently (more than I've published in total). I'm sure they have close to zero downloads (why would they? they're brand new, solve my own problems, I'm not interested in marketing them or supporting them, they're just shared because they might be useful to others) so they wouldn't show up in their review. 2 of these are pretty meaty projects that would have taken weeks if not months of work but instead have been largely just built over a weekend or a few days. I'd say it's not just the speed, but that w/o the lowered effort, these projects just wouldn't ever have crossed the effort/need bar of ever being started.
I've probably coded 50-100X more AI-assisted code that will never go to pypi, even as someone that has released pypi packages before (which already puts me in a tiny minority of programmers, much less regular people that would even think about uploading a pypi project).
For those interested in the scope of the recent projects:
https://pypi.org/project/realitycheck/ - first pypi: Jan 21 - 57K SLoC - "weekend" project that kept growing. It's a framework that leverages agentic coding tools like Codex/Claude Code to do rigorous, systematic analysis of claims, sources, predictions, and argument chains.It has 400+ tests, and does basically everything I want it to do now. The repo has 20 stars and I'd estimate only a handful of people are using it.
https://pypi.org/project/tweetxvault/ - first pypi: Mar 16 - 29K SLoC - another weekend project (followup on a second weekend). This project is a tool for archiving your Twitter/X bookmarks, likes, and tweets into a local db, with support for importing from archives and letting you search through them. I actually found 3 or 4 other AI-coded projects that didn't do quite what I wanted so it I built my own. This repo has 4 stars, although a friend submitted a PR and mentioned it solved exactly their problem and saved them from having to build it themselves, so that was nice and justifies publishing for me.
https://pypi.org/project/batterylog/ - first pypi: Mar 22 - 857 SLoC - this project is actually something I wrote (and have been using daily) 3-4 years ago, but never bothered to properly package up - it tracks how much battery is drained by your laptop when asleep and it's basically the bare minimum script/installer to be useful. I never bothered to package it up b/c quite frankly, manual pypi releases are enough of a PITA to not bother, but LLMs now basically make it a matter of saying "cut a release," so when I wanted to add a new feature, I packaged it up as well, which I would never have done this otherwise. This repo has 42 stars and a few forks, although probably 0 downloads from pypi.
(I've spent the past couple years heavily using AI-assisted workflows, and only in the past few months (post Opus 4.6, GPT-5.2) would I have even considered AI tools reliable enough to consider trusting them to push new packages to pypi.)
This is so stupid. I don't know whether AI has improved things but this is clearly cope, we're not even a year into the transition since agentic coding took over so any data you gather now is not the full story.
But people are desperate for data right? Desperate to prove that AI hasn't done shit.
Maybe. But this much is true. If AI keeps improving and if the trendline keeps going, we're not going to need data to prove something equivalent to the ground existing.
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[flagged]
[dead]
[dead]
[dead]
[dead]
[dead]
This is such copium for AI haters. I stopped working almost any single line of code at the beginning of this year and I've shipped 3 production projects that would have taken months or years to build by hand in a matter of days.
Except none of them are open source so they don't show up in this article's metrics.
But it's fine. Keep your head in the sand. It doesn't change the once in a lifetime shift we are currently experiencing.
No one needs another SaaS. Games are the real killer app for AI. Hear me out.
I've wanted to make video games forever. It's fun, and scratches an itch that no other kind of programming does. But making a game is a mountain of work that is almost completely unassailable for an individual in their free time. The sheer volume of assets to be created stops anything from ever being more than a silly little demo. Now, with Gemini 3.1, I can build an asset pipeline that generates an entire game's worth of graphics in minutes, and actually be able to build a game. And the assets are good. With the right prompting and pipeline, Gemini can now easily generate extremely high quality 2d assets with consistent art direction and perfect prompt adherence. It's not about asking AI to make a game for you, it's about enabling an individual to finally be able to realize their vision without having to resort to generic premade asset libraries.
All apps you are using are made with AI.
Not all of us get addicted to the rat race and wake up at 3am to run more Ralph loops. Some are perfectly content getting the same amount of work done as before, just with less investment of time and effort.
This article is very poorly researched and reasoned, but it's in the "AI hater" category so I guess it's no surprise it's on the front page.
Number of iOS apps has exploded since ChatGPT came out, according to Sensor Tower: https://i.imgur.com/TOlazzk.png
Furthermore, most productivity gains will be in private repos, either in a work setting or individuals' personal projects.