* I waited for Duolingo to support Mongolian for years, and in one evening, created my own DuoMongo https://github.com/matusfaro/DuoMongo
* I wanted an LLM to mock up a server to speak arbitrary network protocols across any layer (e.g. http/tcp/bgp/snmp/arp) for protocol development/testing/cybersecurity and I got it: https://smotanacom.github.io/netget/
* I wanted a way to inspect/route LLM calls coming from my laptop, something like OpenRouter, but locally running, so I built LocalRouter: https://localrouter.ai/
No one is using these, but I love them!
Also it allows me to prototype so quickly, currently had an idea on a Language Server Protocol (LSP) that compiles English language into a set of software requirements (Intermediary Results) and to generate code and tests against those reqs.
I think LLMs are pretty similar to DuoLingo. It is a tool that makes you feel like you have accomplished something without you putting in the work you need to actually accomplish anything. Worse still, you still need to put in the time. So in effect you end up wasting a bunch of time without learning anything new. These machines, are dopamine generators, kind of like cigarettes, or slot machines.
Thariq wrote "personal software was a bit early in 2020 but in 2026, it really can be as personal as a home cooked meal, or a handwritten letter."
Ugh. The most depressing thing about the AI boom is watching tech people devalue human experience.
I recall at the time (late 90s) these were popular dev platforms for regular folks. People would create fitness apps, time trackers, meal planners, etc. using WISYWIG interfaces and visual schema editors.
Of course, the 2000s, 2010s and 2020s were massive consolidation events with winner-takes-all SaaS eating the world.
My intuition here is that software for one is a temporary thing. Longevity has a quality of it's own, and eventually people just don't want to debug their own stuff. Sooner or later $5/month (or freemium) takes over.
I've since tested that theory on the Google Play Store (iOS has one built in), and it was depressingly true. The slow part was trying to find an app that didn't waste your time with ads.
Thoses marketplaces are filled with cheap softwares made to make a quick buck by any means without a thought for the whole user experience, with no way to flags grifters for a better community driven experience.
The economy must thrive I guess.
const audioCtx = new AudioContext();
const sr = audioCtx.sampleRate;
const buffer = audioCtx.createBuffer(2, 2 * sr, sr);
for (let channel = 0; channel < buffer.numberOfChannels; channel += 1) {
const data = buffer.getChannelData(channel);
for (let frame = 0; frame < buffer.length; frame += 1) {
data[frame] = 2 * Math.random() - 1
}
}
const source = audioCtx.createBufferSource();
source.buffer = buffer;
source.loop = true;
source.connect(audioCtx.destination);
const button = document.createElement("button");
button.append("Play");
button.addEventListener("click", () => {
source.start();
});
document.body.append(button);Early home computers would boot up with BASIC. Granted, the user experience was horrible, but at least you could start programming right away.
What is stopping us from creating an operating system that has programming as a first class citizen?
There is JavaScript in the browser, but that still feels as an afterthought, and incurs all kinds of security problems.
Why must we wait for AI to work around this? Is it really true that only very few people are interested in programming? Why not teach it in primary school, along with reading and writing?
Isn't that just any computer booting into a terminal?
in context/mail:
when today/evening:
if content excites me:
send alert
> Our sleep consultant's plan arrived as a PDF full of conditional logic: wake windows, nap caps, what to do when a nap fails
Every so often, you get a glimpse into a whole world you never knew existed.
I had no idea that people did this. This sounds expensive.
Just like other people have no idea that we (Scandinavia) let babies sleep outside in a baby carriage wrapped in a tiny sleeping bag. Fresh air is awesome. Cold air is even better.
This works for some babies and is not universal in any way.
I kinda doubt it's something widespread.
But do doubt away. :)
Please see the official guidelines: https://www.sst.dk/vidensbase/graviditet-og-smaaboern/barnet...
My children slept outside while it was below zero C. The only time they were not sleeping outside was while they were ill.. having a cold or something.
Edit: the limit is -10 C according to https://www.sst.dk/vidensbase/graviditet-og-smaaboern/barnet...
This is well known around here:
it is insane and unnatural but it is also common
fundamentally capitalist concept
contact napping to beyond 1yo is deeply natural and special
Like most soviet children of the 80s, I was outsourced to my kindergarden each mom's workday.
I abolutely loathed everything about it, but the most fascist thing to my infant mind was the post-afternoon nap.
For the life of me, I just couldn't make myself fall asleep, and had to pretend, making extraordinary efforts to really look like sleeping.
We were literally scanned crib by crib, aisle by aisle by the surveilling nanny to detect sleep pretenders.
Those offenders of Morpheus who were caught red-handed, were routinely ordered to step out in front of the entire dorm, which of course was instantly awake and gleefully staring through eyelashes now, and do 20 humiliating squats.
Needless to say, that resulted in nothing but fierce hatred towards the soviet system, which treated children as future soldiers, and oppresion in general.
So let me disagree with you on day napping being a fundamentally capitalist thing, haha :)
The only difference is that it plants the seeds of hatred towards different entities - either the system or the parents.
I now have 3 VPS on Hetzner that run about $14/month in total (the old pricing, thankfully). I mostly got three so I wouldn't worry about wiping one if something terrible happens.
Works like a charm. Even with the new pricing, Hetzner or DO would be cheaper than what OP uses. And the maintenance isn't that hard.
Easier to leave vulnerabilities but you can also do it on Vercel and Neon.
Hosted services are only great for setup and scaling. You're not going to need scaling. Infra setup can now be done easily with AI and terraform.
I spend $2/month on my VPS, write my code in HTML/CSS/Javascript with Posgresql along with other Unix (FreeBSD) tools and the standard languages and get similar things done.
It still baffles me why the heck people pay for services line neon and supabase as well. Utter madness imo. If you add that on to something to get rid of OP could've saved $540 per year.
Pretty much every app I can think of building for my family could just be a website iff local browser-resident persistent data was a thing. (And in some cases I wouldn't even need that if they were going to be always-online.)
As for the banking and identity apps, I probably shouldn't be keeping them on my big cool phone that might eventually get stolen.
Part of my journey has been building out a platform for my apps to live on. Most web frameworks don’t have enough opinionated machinery built into them. You need user authentication, permissioning, MCP endpoints, database access, easy deployment. All of your apps could use this infrastructure and it’s what lets you focus on the actual fun stuff.
I’ve been working on [Homestead](myhomestead.dev) as my OSS solution to all of that.
It's basically an adaptive timer for remaining oneself to be mindful and aware throughout the day, which is a prerequisite for my main practice. When it rings, it asks you if you were mindful since the last timer and adapts the next interval based on your answer.
I've wanted to do this for a while, but a physical device is beyond my skills and I never wanted to bother to learn mobile development. So having the computer take care of all the nonsense was ideal. Very different experience using LLMs when you don't know the technology, but all the nonsense behind creating and publishing mobile apps did not inspire me to learn it.
and i love this tool
That's the result of poor quality, how is wasting time redoing an app mitigate itself?
Programmers want you to review and own the code or they'll dismiss it as slop. Then the lawyers come in and say that doesn't give you copyright over the result. Damned if you do, damned if you don't.
Maybe it's better to just stop publishing stuff. If everything is uncopyrightable slop, the rational move is to just keep it to ourselves. So much for the free and open source software commons.
https://stepwise.run/ Portable workflow orchestration for agents and humans
https://github.com/zackham/aloop Embeddable Python agent loop for integrating LLM agents into your projects. Multi-provider (OpenRouter, OpenAI, Anthropic, Google, Groq), built-in tools, hooks, ACP server, and persistent sessions.
https://github.com/zackham/plat An embeddable infinite canvas to build on. Bring your own node types; drag, zoom, edges, auto-layout, and undo come free. Host-agnostic storage, runtime API built for agents.
https://github.com/zackham/seance Shared live workspace for humans and agents — multi-pane terminals, notes, and a control plane for co-working in the open.
deep dive docs for my personal agent system i've been developing since dec, and using constantly: https://vita-docs.ham.xyz/overview/
browser replacement for stuff i control: https://vita-reports.ham.xyz/s/2119130b0b6b
a snapshot in time of the education system i run for my kids: https://vita-reports.ham.xyz/s/e7036ae931a7
if you want to talk about any of this stuff, or have a related project you think i'd appreciate, hit me up zackham@gmail.com or @zackham on telegram. abundance is here for personal software. take advantage of it and have fun!
I'm the sort of person who leans on Postgres instead of sqlite, but I'll put aside the sqlite argument for now because the fruit is so low hanging here. These use cases could probably work just fine with something far less robust. A redundant cluster of two db.t4g.micro RDS instances is ~$20/mo. But you probably don't even need that. Even just a container with a persistent volume seems like it would do the trick. With a little bit of caching, all of this could probably fit in the AWS DSQL free tier. And it brings me no joy to say that using AWS managed offerings is the operationally cheaper option.
I love Neon and think it's an incredible service, but this is some Heroku-ass pricing.
I'm not sure I believe this is a real job. Do we really need to pay consultants to tell us that babies need naps?
We managed the first two so we are not completely incompetent:) so no it turned out very useful to have a "sleep coach" give us strategies that we hadn't thought of.
Nothing as elaborate as what's above but not completely ridiculous.
Another observation : I expect in future the UIs could be dynamic, fluid, AI-generated and personalized. Old person, here's big font for you. You need to just know current balance, here's the number.
About four years ago I started building an RSS reader that was supposed to use an algorithm. The problem is, I didn't code in years and wasn't familiar with the modern web stack. I had to drop it, although was still able to sell the domain I got for $500 (nice).
And then, with Codex and Claude, I finally built it this year and enjoy it every day, along with 30 other strangers, since I made it public. It's a great feeling.
I now finally have an app that I use to save both links and articles to read, sort of in-between of Readwise Reader and Raindrop (and it does sync with Readwise).
I write my blog posts in a little browser-based note-taking app with sharing. It's minimal and looks great but has all I need.
I built a few Chrome extensions that I really wanted but couldn't find (well, couldn't find beautiful options).
I don't think everyone will do this. And getting the audience is a major pain, always has been, but now most of your early adopters just go and build their own things just like you. I'll still take it.
There's one in particular that I use quite often and have for about a year, vibed for myself: it's a chat interface that walks you through processing an emotional or difficult moment, following a process / workflow. Supports Martin Seligman's ABCDE and Byron Katie's "The Work". Two techniques that I found most useful to improve my thinking and responses to difficult situations. You converse with it, and it leads you through the stages of the selected flow. More than a system prompt - it tracks and progresses through the workflow at the right times, and you download a consistent pdf of key details from the "session" at the end for your records.
The thing is, I can't sell it. I don't even think I can open source it. It's mental health (minefield of legal and ethical issues), and would probably breach copyright, trademark etc law.
Yet it's very useful to me. I guess it's the equivalent of making a system to do these things at home, like prompt cards or key points taken from the books, with a well formatted note taking structure. Except it's way easier because you just chat like you're talking to someone.
I would never have spent time building this. But now it's so cheap and easy to do.
If you freelance (design, illustration, photography, writing, or solo dev): which part of that would you most like to never think about again, and what did you try that didn't stick? Genuinely curious rather than pitching anything — I don't have a product, I'm trying to figure out whether the real pain is the bookkeeping itself or the decision-making around it (pricing, when to chase, what to set aside).
Then I did something similar for my son, giving him a targeted job search app specific to his Gen-Z brain responds to, again launched off Github pages as a PWA, but this time with WebRTC to allow iPhone to laptop data syncing.
And along the way I realised there's probably whole host of people for whom software has never really worked. And I can built them things on a highly personalized basis that have only minimal reliance on cloud-based services.
And then I went a bit crazy and created a manifesto which I've done nothing else with https://harryf.github.io/reclaim/ ;)
Basically the agency I work pays every two weeks and the time between I receive an assignment and the the time i get paid for that assignment is about 45minutes.
I built an automation, that when I get a new assignment it automatically adds the assignment to my calendar, to the spreadsheet and it calculates the mileage using google maps API.
Then the spreadsheet calculates how much I’m gonna get in the following 2 paychecks and the one that’s currently running.
* Doom launcher and Megawad Steam-like library for macOS: https://github.com/stared/rusted-doom-launcher
* TRMNL X e-ink screen vibe coding usage screen: https://github.com/stared/trmnl-x-claude-codex
* And weather in Warsaw: https://github.com/stared/trmnl-x-weather-warsaw (here's location is hardcoded, so the expected use is fork and vive code to match both your location and visual preferences)
* Image upscaling script for a thermal camera, https://github.com/stared/thermal-upscale
And many, many others I did not share, mostly because they were too peculiar, really unlikely to be useful for anyone else than myself.
Writing code with one user is the base case of a “why am I writing this” recursion - because I use it.
Every action leads to next insight. For me, a browser extension that provides a table of content to chat interfaces because I always need to scroll up and down to read preview queries. Writing browser extension leads me to writing a RSS reader navigating with only keyboard. Writing RSS reader leads me to writing a virtual book shelf that visualize Goodreads record on a virtual 3D bookshelf. Writing virtual book shelf lead me to making a virtual climbing gym where I scan my local climbing gym into 3D model where I can map climbing videos on it. And it is just one chain
There's existing stuff for this but making a personal extension to this should be very well possible too.
Too many marketing persons.
Let them talk to themselves
At work the team is leaning be heavily into internal tooling. Doesn’t need heavy support. Doesn’t need a fancy designer. Duct tape glue and liquid wrench are more than enough. It’s quick enough that it’s throw away and so if need be we can always write a v2 without losing months or years of work. Just enough to be useful, and then back to actual work.
Of the two, iOS is /worst/ for running vibe coded apps.
You need to pay $99 a year to be an Apple developer if you want your app to live on your phone for more than a a week.
Distribution is a also pain in the ass even for personal use alone (you can't just toss around an IPA).
Finally, so many useful APIs are totally locked up. The fact that I cannot in 2026 extend Siri, for example, us infuriating.
Android is far more promising but Google recently shot themselves in the foot with their verified APK nonsense. Google's privacy posture is infamously nonexistent as well, and I have little interest in rooting a Pixel at my age.
Hopefully now that enough people can vibe code, these walls will crumble a bit.
https://apps.apple.com/us/app/delta-chat/id1459523234
The beauty of the ecosystem is that apps are just zips of a webapp that is self-contained, and uses the webxdc api to message other copies of itself running on other folks' phones that are part of that chat group. Basically, the app communicates (and is distributed over!) the group chat (it's otherwise insulated from the internet). And because apps are just zipped webapps, they can be pulled apart, hacked on, and reassembled as needed for your group.
ChatMail relays tend not to retain info - they aren't hosts, they just relay encrypted messages (the platform mandates encryption). This sidesteps all the approval/review processes, and there is no app store or developer fee or even SDK. It's just folks trading apps and running them together.
It can't solve a lot of cases, but it solves some use cases (like a shared grocery list, or real-time card games, or games with leaderboards) really well.
My son had written a game like this called "Quantum Chain" that had basic leaderboard functionality. That was the first app we ported: the platform injects the webcdx.js shim, and we update Quantum Chain to post messages to it when someone finished a game with a high score. This is replicated out to the relays the same way every message in the group chat is. When the app fires up, Delta/ArcaneChat replay all messages that are typed for that app back to it, and it restores it's state that way. There's some localstorage caching to make this a bit more efficient as the stream grows.
So it's a bit like several services reading off the same kafka stream to stay in sync.
I've since ported over (and massive improved) my implementation of KenKen, which also has leaderboards. I then took on realtime gameplay: a card game my family likes called "Shithead" that supports 2-6 players. I built a first version in about 6 hours, and I'm still working out session management (so folks can join and leave games and everything syncs properly). Still a couple of edge cases to fix, but overall, it been trivially easy to bring over webapps to the platform that already exist, and a few hours of effort to create a new one.
I'm pretty thrilled with the whole ecosystem, as it gets me away from all the BS the companies have layered on top of the basic "Apps with Friends" experience ArcaneChat offers.
You can make perfectly good apps that run great with a free account. They expire in a week instead of 3 months (for TestFlight builds). That's it. You can redeploy with minimal effort. You have to reauth with Apple every ~30 days. The reauth is the only part of that you can't have a cron job run (you need to authenticate the keychain).
For normie folks, the hoops here are mildly annoying. But if you're someone making vibe-coded apps, we're talking about two scripts and a cron job, here. Or just run Sideloadly, which I hear works pretty well.
There exist reasons to be annoyed by Apple's iOS security posture, but "I can't run my vibecoded apps" isn't an especially good one.
Can I ask what your workflow is for this "minimal effort"? Maybe I'm missing something, as I thought it would require manual faffing with laptop, Xcode, etc. every week?
(PS I've tried the two prominent sideloading approaches, but couldn't get either of them to work on my iPhone.)
No thank you
For personal apps, the locked down stores are making themselves irrelevant. When you have software for one person, you don't want to publish it there. You want to bypass the gatekeepers.
Two apps fill niches that I've wanted to code myself (one, I've wanted to code for the last twenty years on and off), but have never had the time to do so. Now, the speed from thought to working prototype is insane (less than an hour); it then takes a few days of occasional work to turn that prototype into something architecturally sound and half-decent.
And after that... well, I just keep improving, refactoring, refining, because it's so much fun.
But yes, I don't really want to pay Apple $100 a year just to run my apps. So I just have them as PWAs. Wouldn't work for many use cases, like when you need Apple Fitness data, but works for me and it looks fine.
Don't do that, that's a PITA. Put GrapheneOS on it instead—simple, no hassle, works great
No complaints over paying more than >$100 a month in subscriptions with coding agents but $99 a year is some how too much?
Yet another reason why “developers” are just consumers that are selective complainers that want their tools to cost $0.00.
Clearly as already stated in the article, hosting and distribution is not free.
If I write an app for myself there is no need for hosting or distribution - it costs Apple absolutely nothing, and they don't need to be in the loop.
I've been able to run my own software on my computers forever, and still can on Windows, Mac, and Linux. It's just the mobile OSs that are different, and they shouldn't be. That's what sucks.
---
Related, I've been wondering if this issue will hit more and more consumers as vibe-coding personal apps becomes available to more and more people. Apple may have to change this aspect of its model, in time.
Plenty. The problem isn't so much the price as it is the loss of sovereignty.
It's not your model running on your computer, it's $CORPORATION's model running on their own data centers. They're just generously allowing you to use it, and only on their terms.
Just discovered webxdc in another comment here! https://delta.chat/en/2025-01-23-webxdc-no-billionaires
Aside from Apple, running homemade software on a phone was free of cost, even back in the days of Motorola flip phones running Java…
https://blog.moralestapia.com/posts/application-paradigms-en...
LLMs are a good help here because side projects typically slow down for me when I run into bugs / troubleshooting. So while I genuinely believe struggle is necessary for learning, being able to get over that hump to build something I need makes a big difference.
I'm not vibe coding the app though, since I actually want to know what it does. But I can see it being helpful for people who just need something and don't care about how it works.
But it is! The Patron level of Pocket Casts is $15.99 / month!
I used to love Pocket Casts, but it didn't have the ability to easily add an RSS feed URL locally without submitting it to their index - which I don't want to do with my AI's personalised morning podcast (and that's another category, Podcast For One). I've had to use AntennaPod for that, but it's missing all the silence clipping and audio normalising features that Pocket Casts has. I'd love to see someone improve on it, and I've been tempted to do it myself too. Good luck!
Now I'm actually doing it.
I've had an idea for how I want to be social with friends and family (a website, but setup within my vision). Now it's done
Could I have done all of these things on my own or just paid for existing services? Yes
But now AI can do it for me while I watch TV. I just show up to test.
But isn't it like super cheap to host a website in 2026? A lot of those builders are actual better if you just want to write. Otherwise you will have to maintain your website and also focus on writing, which would not allow you to just focus on writing.
I am not saying this out of my behind. I have done the exact same thing and ran into issues. If you just want to focus on content, building and maintain your own website can be a big distraction
> A fitness app that sizes my smoothie to that morning's run
I don't feel I need more than the phone's calculator app for this particular problem.
Other examples could be solved by the Notes app, I feel.
I deliberately avoid creating new "apps", and make do with existing apps and solutions and create workflows rather than apps.
I write scripts, add features to scripts and understand all lines of them.
There are very few exceptions to this, like, I vibe coded a visual PDF cropper app that also handles scaling, merging in a GUI built with PyQt and GTK.
Having entire apps create maintenance burden and headaches.
Is the article interesting/ relevant or not?
If not, doesn't matter if it's ai slop or human slop.
If yes the. It's good that the author could use AI to get it out vs not writing it.
Your app tracks calorie intake, doesn't keep track of total calorie burn of the day (maintenance calorie tracking)
And no, calorie burn from watch is not good enough!
>A 2020 systematic review by Fuller et al. found that consumer wearables often perform poorly at estimating energy expenditure, even when they measure steps and heart rate reasonably well. Across brands like Apple, Garmin, Polar, and Withings, calorie burn estimates frequently showed substantial error, often in the 15–40% range and sometimes exceeding 50%. Wearables can still be useful for tracking trends, but not for precise calorie calculations.
Here's completely free one i built, https://macrocodex.app/ 16,000+ users already.
The actual method used in the app is a bit more complex than this, but I think this explanation is good enough to convey the core idea for now. Later, I plan to publish a more detailed write up.
>Also, how much better it is? How do we even know it’s better?
Very simple! If a car salesman tells you that they use space technology in the engine to improve fuel efficiency, how do you know whether they are telling the truth?
You fill the tank with a specific amount of fuel and drive. If the fuel lasts much longer, you know the technology is actually improving efficiency.
The same idea applies to MacroCodex.
When you eat at maintenance you do not gain weight, when you eat above maintenance you gain weight, when you eat below it, you lose weight.
If you set a goal in the app to gain 1 lb per week and consistently eat the calorie target prescribed by the app, then your actual weight gain should match that rate. If it does, you know the maintenance calories calculated by the app are accurate.
You can choose to set "maintain" or "lose" or "gain".
Not for me.
I like software for people, and in terms of affordances and utility mass buy-in and standardization is part of the point.
I can invent a new kind of key. I could…
Maggie Appleton said that it would enable "barefoot doctors" to bring helpful software to small communities. I have seen a lot of those underserved groups in my dev career.
I am not too happy about AI killing my business and inserting slop into every aspect of my life, but I am very hopeful about that aspect.
---
I also made some apps for one in the pre-AI days, and now I can quickly add features I've been procrastinating for ages. I can finally process data that I've been collecting but had no time to write code about.
Maybe Google is eliminating sideloading to force people to not make their own apk’s now that it’s easy to do so.
Hasn’t cost me anything.
I have written so much software for one... I cannot recall everything. I use some of it daily:
- I have my custom elf(glibc)/linux distro
- I use my own ffmpeg based x11 media player (will have an update for wayland).
- Since the "geniuses" at gogol blocked all accounts not using whatwg cartel web engines and blocked all self-hosted SMTP servers (I do not pay the DNS mob, then IP literals, which is stronger than SPF) to exchange with gmail.com users (now prisoners), I have my very own simple noscript/basic HTML servers (messaging, file transfer, openstreetmap browsing, etc). Unfortunately, most are C written from linux instead of assembly (at least, I removed most of the time the libc dependency with direct syscall programming). IPv6 is making everything internet software muuuuuuch easier to implement: I am lurking at an IPv6 only, super idiotic and real time voice/video/text/file transfer protocol, probably based on SIP (the main issue are randomly generated IPv6 addresses from mobile internet, which may require a mini-server for IPv6 address exchange, BAAAAAAAD!).
- Ofc, I still have my own minimal SMTP server (and a SMTP client to send emails), because there are still honnest people on internet. mutt forever.
- I am trying to build a RCS for linux, aka a Reduced Command Set with near direct hardwiring to linux syscalls. I lose comfort, but knowing that I removed giga tons of bloat feels so much satisfying it is worth it by light years. I will have to code my own command shell one day.
I still have C code since I currently need portability on IP-LOCKED ISAs (arm and x86-64). But I wish for RISC-V to be a success, namely as a modern, very often hitting the 'sweet spot' in technical design compromises, NON-IP-LOCKED ISA. I am writting much software in RISC-V assembly now... which I run on x86-64 with a small interpreter (written in x86-64 assembly)... trying to fix the bloat and kludge from corpo-like open source software (including the SDK).
Right now, I am writting my own wayland compositor using that framework. I was not expecting window management to be a bazillion of little things to do everywhere. And even if a wayland compositor is several orders of magnitude smaller and leaner than a x11 server, it ain't that a small project (ofc, I implemented myself the wayland wire protocol, no external libs here).
In my 'trying to fix the bloat and kludge' from corpo-like open source, I force myself to use my own exe/dynamic lib format for modern hardware (so lean, a simple RFC will be enough, and no more loader bloat/kludge). But I am scared at the abominations which are all the requirements of classic computer languages to run: I would like a mesa vulkan driver, but there is still c++ (super BAD) in there which makes porting to an alternative dynamic lib format a living hell, non trivial C won't be that easy neither (for instance the abomination of the ISO __thread keyword, amazing bright minds there). Taking perspective from that, I wonder why anybody sane would add on top of that a big runtime requirement??
And what people doing that have to keep in mind: IRL can put a violent stop to all of this.
I will if you are building anything good. What even is this statement?
You'll just prompt your LLM provider and they'll provision your software. It'll do testing, dependency updates, monitoring, and regular audits.
If you want fixes/features, you'll prompt your LLM and it'll make it so.
I've been using the agents since the day they came out. This, does not work. It works if you want an app that looks like it works if that works without any testing. If you use the app in any way that isn't boiler plate it will break. Because you can't prompt ai with a simple feature no matter how small and have it cover all the important cases.
What these models do is the absolute bear minimum to satisfy the user even if it means cheating. And anyone who doesn't look thinks its miraculous. It really is not.
So, you absolutely can prompt it to add a feature, but you need to prompt it 10-20 times to get the actual feature built. If you don't know this yet rip.
And before we continue, also known that when you prompt for the next feature. There is a high chance the last feature you built will be altered or deleted.
Opus 4.5 with a plan mode is enough for me: it asks clarifying questions to map out the design space, it covers fiddly parts, it generates good unit/integration tests.
I have seen it cheat by deleting existing tests, since it's single-mindedly focused on the problem I give it. I've not used an LLM-to-review-an-LLM, but when I've used an LLM to review my own work, it notices when there are unrelated changes in my commit/working copy.
Besides, all of your critique is how I've heard product managers speak of human developers! And a tool needn't be perfect for it to be useful.
* A personal music player for Apple Music with an embedded agent loop, so I can say things like "I'm going to work in the wood shop, play woodworking-appropriate music and keep the playlist full and non-repeating indefinitely".
* A calorie/protein tracker that takes 1-liner summaries of meals, resolves them to macros with GPT5 calls, and tracks them in SQLite.
* A macOS menubar combined Apple TV and Roku TV remote control.
* A calculator front-end to SageMath that renders Latex math notation and has point/click shortcut affordances for linear algebra and MVC.
* A "self driving wiki" macOS desktop llmwiki that imports and digests PDFs.
* A menubar tracker for the little temperature sensors I have scattered all throughout the house so I can click my menubar and instantly see the temps in the different rooms of the house.
* A layout tool for designing picture frames to get cut lists and rabbet depths and stuff.
That's just the desktop stuff. These applications are generally better than what I'd get from the App Store. The calorie tracker is one of the better-designed apps on my dock.
I feel like I'm not doing this enough yet, it hasn't fully sunk in, and I'm going to be building a lot more stuff when it does sink in. I haven't yet gotten to the point of writing my own note-taking application (still Bear.app for that), but how long can that last?
In the past month I’ve completed a home dashboard idea I’ve had for years, I’ve built a house management system, a travel planning tool, a guitar chord annotator… it’s been very liberating. I’m not doing the coding, which was fun, but I am doing all of the system design and iteration, which I also really enjoy.
The other day I refused to fire it up for him and he had a full meltdown. I told my wife I had never seen product market fit like that before.
It is a little harder with something like a game to maintain motivation and go for perfection when you know your audience is an audience of one but it’s also a relief to just worry about the things he cares about (trains and firetrucks).
I also miss there being something intellectually challenging to a side project but that probably means I just need to learn a real game engine after this.
In the agentic era, there is no need for maintenance - each artifact is immutable and just-in-time. In case you need a new version, spin up a new artifact.
I’m only half-joking.
Like one of the most obvious things that a computer can do is run a task periodically at a time to go maybe fetch information and process it. And yet, the average computer user using any commercial operating system does not understand how to set up a recurring task. Operating systems don't really advertise it or talk about it as a feature. I think this is a good example of the disempowerment of end users by software companies.
LLMs have the potential to change this by allowing people to simply ask their LLM to invoke the right incantation to schedule the task or produce the program. That seems like a great change. I'd like to see more power in the hands of the end users rather than less.
For people that don't want to put in the effort, of course, this will change nothing. But for those that are willing to explore a little bit, I think it can be a big improvement over the software ecosystem that's emerged in the era of the app store.
The "run a task periodically" isn't the difficult part, it's processing the data that is.
And to be honest I don't think llms will empower that many people, the vast majority don't want to put the effort required (and it's definitely more effort than writing a one-shot prompt)
I wasn't really talking about fetching email.
Or with voice. Or even face movement control, like via the camera.
how is AI (or a human for that matter) supposed to know what “woodworking-appropriate music is”? I had the same thoughts when spotify released their “let AI make a playlist for you”. Like, we already have recommender algorithms, why do we need to get a LLM involved?Maybe I just don’t understand woodworking and the connection it has to music though (I’m also the kind of person who’s had the same playlist for like the past 4 years)
But, I do want to say, personal app built just the way you like it, an alternative front end to a closed source ecosystem even, is super cool, and has been one of the things that AI coding has really enabled.
I’ve done a few project where I just give claude a har file and tell it to reverse engineer a cli from some web interface and it’s so powerful to be able to do stuff like that now
It doesn't, but you'll believe it is because that's what you asked for. It's not like you have a solid test for what it is yourself
Thomas explained what he means in a sibling comment an hour before yours.
I would probably think about doing a little workspace sharing and ask for more live musicians to come in and do some serious woodshedding themselves ;)
I know some that are getting that buzz-saw sound out of their instruments fairly repetitively already, and there's no place to go but up :)
To be fair, they’re also easy (trivial, even) without one. It’s not like “play more music like this” is something we ever needed AI for.
Borderline off-topic, but I love how fast the goal post moves. A decade or two ago "Similar music to this" was definitively in the realm of AI/ML, now it's some full-stack dev's afternoon to implement.
Of course not, but just some years ago, when people talked about "AI" in ML circles, they weren't exclusively talking about LLMs, which you seemed to have automatically and exclusively associated "AI" with. It's much larger than just LLMs, although the ecosystem is flooded with LLMs at the moment. I thought this very movement, where tons of people, not just you, have this "AI === LLMs" association, when again, it used to not be so.
That attitude is exactly why online (and offline) discourse has gone in the gutter. Instead of addressing what was said and understanding nuance, you instead took the comment, labeled it in a preconceived box in your head, then attacked that group instead of engaging in good faith with the person you were talking to.
And by the way, I’m very much aware of the difference between AI and LLM, but whether you like it or not the terms are conflated today and being pedantic about it is a losing battle. In this case it was patently obvious what was meant. You’re (presumably) not an automaton and are thus (presumably) capable of understanding the difference, meaning you are definitely able to choose to steel man instead of straw manning.
What on earth are you on about?
Take a breath, I'm not arguing for/against anything at all, I conceded myself it's "borderline off-topic", you don't have to take it as I'm trying to challenge anyone's worldview or anything, it's basically an offhand observation.
> man instead of straw manning.
You may not actually understand what a "straw man" is, as I haven't even attempted to refute or engage in any argumentation at all.
Anyway, clearly a fruitless conversation to continue, I'm not sure who you're arguing against/for here, I didn't intended to even try to argue against/for anything here, so I'm just gonna peace out. Hope you enjoy the rest of your Saturday as much as I've enjoyed mine so far!
Here’s a silly example: Imagine you had made a comment about something different on a different thread. Then someone responds to you by saying “borderline off-topic, but I find some comments on HN about this subject to be really stupid”. Would you not be justified in thinking that the comment responding to yours was a response to you?
Maybe you really did mean nothing by it, but are you able to understand how every choice you made contradicts that, even if inadvertently?
> Anyway, clearly a fruitless conversation to continue
Are you also able to understand how this type of high-ground dismissal makes it worse?
> Hope you enjoy the rest of your Saturday
Thank you. You too. Genuinely.
I have a bunch of skills Ive created for random work that have “predefined interfaces” like a single url. I very frequently trigger these skills with freetext prompts instead and get all the intelligence of the skill while bypassing what it was actually designed for. This aspect of malleable API is extremely powerful and is probably the only reason I keep AI in the loop of some of my custom software vs removing it and avoiding the cost/nondeterminism.
It’d be like eating and having an alien with no concept of food or taste staring at how much your facial muscles change and making notes. Then recommending places to you based on the other people they’ve been studying.
> how is AI (or a human for that matter) supposed to know what “woodworking-appropriate music is”?
Well someone is going to be concentrating, but it’s a long period, with loud activity. maybe you’d guess at decently active music, not relying on careful complex listening, probably fairly steady rather than sudden dramatic changes.
Maybe it could save previous playlists and what you’ve said about them, or ask questions like “so like, really calm or techno or what?”. Adding just a bit of common sense makes an enormous difference with this kind of thing.
(Actually I totally do want to allocate lots of time curating exactly what I'm going to listen to, but laying out playlists is its own hobby, and woodworking time is for making wood shoot across my basement when I pinch the saw blade, not for making mixtapes.)
“If I Had a Hammer”
“Sledgehammer”—Peter Gabriel
Anything by The Carpenters
Any works by Woody Guthrie
Hymns about Jesus and St. Joseph
Lumber Capital Log Yard
Anything by Nine Inch Nails
Anything by the Sandpipers
“Turn, Turn, Turn”
“Dust in the Wind” by Kansas
Anything by Dusty Springfield
Anything by MC Hammer
Original Motion Picture Soundtrack, “Saw” horror franchise
Anything by Level 42
https://youtu.be/dQw4w9WgXcQ?si=sTUKhRXXiylavwIm
Anything by 1-hit wonder The Nails and/or powerviolence band NAILS
“Morning Wood” by Beavis & Butt-Head
“The Lumberjack Song” by Monty Python
and we had autocomplete for coding (IntelliSense), why do we need to get LLM involved?
LLMs are at a whole different level than recommender algorithms
tried it with GPT:
> A steady mix of instrumental grooves, classic rock, and bluegrass—styles commonly favored in woodshops.
Green Onions - Booker T. & the M.G.’s
Up Around the Bend - Creedence Clearwater Revival
Jessica - The Allman Brothers Band
Long Train Runnin - The Doobie Brothers
Honey on My Tongue - Steep Canyon Rangers
Indeed.
I'm not sure when, but at some point after I saw Claude Code, I realised what I wanted for my next OS was basically an LCARS-themed* thing, where the "desktop" is a TUI and whenever I needed an app I didn't already have, I'd ask an LLM to write it for me instead of downloading it (benefit: the app would never have an update pending (which might break my flow) that I had to install before use, unless I'd specifically asked for a new feature to be written); and if I wanted to browse something on the web, the browser would use something like an LLM to strip out everything that wasn't content.
LLMs aren't quite good enough for that yet. They're interesting, I've also made a bunch of apps this way that was an easier experience than even figuring out which offering on the app store was right for me**, but still, the LLMs are not quite good enough; I'm pining for the OS equivalent of a car with a self-driving mode so good there's no steering wheel and the windows have black-out shutters so you can sleep while it drives you and your insurance is fine with this.
* because I'm that kind of nerd
** e.g. rhythm action game that takes midi files for the songs and listens for a midi keyboard for your input: https://benwheatley.github.io/Piano-Trainer/piano-trainer.ht...
Your feedback as a builder is greatly appreciated :).
Doctors are not experts in nutrition.
Most modern western diets lack fibre the most so eating more fibre is a sensible recommendation for basically everyone.
Chill with the snark for a second and realize you may not be the smartest person in every room.
His advice is literally “He needs to just Google it.”
I've:
* Checked these against all three major frontier models and gotten the same macros breakdown within relatively tight bands
* Manually researched a couple of the most common recipes
* Run this system over many months against my most likely TDEE and reconciled against scale weight loss (which has been profound).
My confidence in frontier model's ability to generate "good enough, probably more than good enough" from this level of meal description is quite high.
Key simplifying assumption (for the model): I'm cooking most of my meals. I cook. I virtually never order out or eat packaged food of any sort. This would be a harder problem and the answers less reliable if I was giving it, like, "1/2 JP Graziano muffaletta, extra hot giardiniera".
When I reflect on my own life and think about why I'm not personally adopting LLMs for anything, it occurs to me that I just don't feel like I have any problems like this to address. I was already eating appropriately, already working out appropriately, already happy with the music I listen to, already happy with the way I learn and practice math. I could make a billion personalized apps, but I was already a two-time state champion in cross-country and track who has never been overweight, got perfect scores on the SAT and GRE, won regional academic olympics in math, all before mobile apps existed and before I ever owned a personal computer of any kind. That's not to say I can't do even better than I did as a teenager in the 90s and I try to maintain some proficiency in these things into middle age, but it never occurs to me that I need or should need software to help me with this. I have no clue what the temperature is right now of any room in my house, but I know they're all in a comfortable range that isn't hurting any machinery or animal and that seems good enough to me.
I see all these enthusiastic booster posts and comments about all the tools you can make, but I don't feel like what is missing from my life is I don't have enough tools. I wasn't sitting around prior to GPT-3 thinking to myself "you know what would improve my life? More software." There are deficiencies and problems to address for sure, but it's mostly issues of prioritization and motivation. I have fallen off the wagon a bit this year and don't exercise nearly as much as I should. I don't talk to my wife enough. I don't talk to other friends at all. I don't go outside enough. But those aren't problems to address with software, either. If anything, if I spent more time making software, even with LLMs, that would be more time not going outside, talking to my wife, or working out. No matter how much raw output I could get computers to output, more time spent fiddling with computers is not going to improve my life. I shouldn't even be here right now, reading and commenting on Hacker News. The sun will be up and it will be 90 degrees fahrenheit within an hour of that. This is my last chance of the day to go run without feeling extremely uncomfortable while doing so.
This isn't a "oh my God look at me I'm so much better than you" comment or at least that's not what I'm trying to get across. But I'd encourage the original poster here and you as well to be outcome-oriented. My weight isn't in a warning zone. This guy is posting stats, so I am leaner than he is and my marathon time is faster, all without the aid of any tracking apps. You're a public figure, so I know you're in the same time zone as me, which means you were posting your own comment at midnight last night, midnight on a Friday night reading Hacker News. My sleep isn't perfect, but I'm not reading Hacker News at midnight on a Friday. Are all these apps actually making your life better?
No. These people just spend more time fiddling with their computers instead of doing the parts of life that it supposedly frees them up to enjoy. It's IoT lightswitch automations all over again. There's no reason anyone would ever need to calculate their smoothie according to their Strava data and the positions of the planets on that day. It makes far more sense to spend more time knowing your body and learning to treat it well than trying to create an app that will do it for you.
You're looking down on the people at the bottom or middle of the mountain, struggling up hill to reach the peak, and wondering "why are they going through all this effort? Why don't they just stroll around on flat ground like I am?".
They're still building the systems that you already have in place. It's far easier to use a system you already have than to build one from scratch. Maybe not with physical or academic health for you, but I'm sure there's some areas of life that you're lacking in that could use improvements. If you can't think of any, I can give you a hint: empathy. If you try to work on it, you might have to do things like set reminders to call friends on their birthdays, or just to check in. You might have to force yourself outside to chat with that neighbor that you don't exactly like and find some common ground. You might have to set a goal of volunteering a certain amount of time at a homeless shelter, and track those hours.
Meanwhile, there are people who already have buffed up their empathy their entire life, and do all these things(and more) naturally.
Should they look at you and wonder why empathy is so hard for you, and tell you your efforts to consciously improve this part of your life is a waste, since they were able to achieve higher levels of empathy without doing any of those things ?
Bull. They're doing highly sophisticated procrastination. I'll never become a great photographer by vibe coding the perfect tool for my workflow that selects, ranks and edits the 100 photos I took this week, or one that tells me where to stand and when to click the shutter. I have to be out shooting thousands of photos every week and laboring over the editing myself. You get good at what you do, and if you're spending your time making these one-off disposable apps instead of doing the thing they ostensibly support, a few years from now that's what you're going to be better at.
> I have to be out shooting thousands of photos every week and laboring over the editing
Why? Ansel adams would only shoot 20 photos a week on average. Why are you procrastinating by spending so much time with the camera instead of just having naturally good taste like him?
What if I just like making my own stuff?
Surely that’s plenty.