Software is a manifestation of someone’s knowledge of and experience in and ideas about how a thing should work. We learn from the software we use, we benefit from everyone else’s ideas, we benefit from the hundreds and thousands of hours other people put into understanding a problem to design a solution. My workflow is better because of the incremental improvements made by developer after developer year after year. Would we have Claude Code if our foredevelopers hadn’t spent thousands of hours deep in thought, obsessing over every last detail?
Building all the software you use yourself, whether by hand or by vibe coding, cuts you off from the world.
I have no philosophical objection to vibe-coding apps for yourself, but personally, I wouldn’t be 1/10th of the engineer I am if I wasn’t constantly exposed to the work of others.
For some, this trend worries software engineers — who needs software if they can vibe code it themselves? — but I am much more optimistic. I think people will start valuing good software a lot more. Claude code can deliver the first 90%, but we all know it is the last 90% that differentiates.
I like to say "my code is 200% vibe-coded; the tricky bit is figuring out which 100% to keep".
Decisions matter, both technical and product ones. LLMs don't make as good technical or product decisions as I would, and the way I work with them tries to maximize my strengths and the LLM's strengths. I don't know if I succeed, but it's better than "make me an app like X" as a prompt.
Your entire post is self selection bias and survivorship bias.
SWE field is one of the most cognitive dissonant social groups; cries foul at the slightest whiff their free speech and agency is being put upon; seeks to reduce blockers to their productivity, fewer PMs! Less management!
Now complains about users using their machines without having to block on an SWE.
Insert that quote about how someone will not see the obvious if their paycheck relies on them ignoring the obvious.
Here come LLMs and all they can accomplish with a few arithmetical rules instead of the arbitrary semantics of an SWE; watch as SWEs block social evolution away from disrupting software engineers.
As an example; "protected memory", among many other individual software problems, is an access control problem mired in old semantics relative to OS monoliths.
Didn’t see you all halting as you decimated travel agent jobs, retail jobs, etc etc. Technology advancement must now stand still after centuries of evolution? The self selection bias is as obvious as Trump's.
Making your own software is a good way to escape enshittification and influence.
I switched from Spotify to buying MP3s and using my own audio client, because I'm fed up of a company telling me which music I should listen to every single time I open the app. It costs more, but I own the music and I escape the constant redesigns, price increases and influential behaviour.
Most apps are very simple and there isn't too much to learn, especially if you're building it to scale to a userbase of yourself. I can't see the need for a ton of CRUD apps which demand subscription fees personally. If you build them yourself, you get to keep your own data, build it out the way you want it, keep it that way, and use computers as a person using a tool as opposed to a customer buying a product.
First of all, I’m skeptical about these being free. Time isn’t free, and the tokens to make these projects certainly weren’t free.
Second of all, all of these SaaS apps that don’t actually have a need for recurring charge probably should be paid one time. I don’t use Loom — I use CleanShot X and it was a one-time $30 payment and has a lot of great features I benefit from. I can’t reimplement it in $30 of tokens or $30 of my time.
But for an app whose use case doesn’t change and is recurring for no reason? Yeah there’s probably not much value in recurring payments outside of wanting to support the developer. I pay a lot of indie devs out of the goodness of my heart, and I’ll continue to do that.
But the value for “SaaS apps” without clear monthly costs should have always been under scrutiny.
Most apps rationale for subscriptions is "Ongoing development" without an option like jetbrains etc. to fall back to a perpetual license.
In practice, regardless of whether an app needs ongoing development or not, this is the best way to try to guarantee continuous income and make a living off of a project I guess.
Perhaps LLM's will force developers/companies to change their stance and to stop users from recreating what they have already created, just buy an at-a-time snapshot of their app for a one-time-fee? Probably not but one can hope.
It is not irrational on the part of the developer -- I've definitely felt this too. The problem comes from the fact that practically everyone has subscription fatigue these days, and each of us probably has only a few pieces of software we truly care about enough to want to support them out of the goodness of our hearts.
But everyone wants us to pay $10/mo. It just isn't sustainable from a consumer perspective.
> But everyone wants us to pay $10/mo. It just isn't sustainable from a consumer perspective.
And so few actually deliver $10/mo worth of value. If 1password and Fastmail - the two most important services that control my digital life - are each $60/year, that's the standard of value other SaaS companies have to beat and very few do. The ones that do are like NextDNS where they cost $20-30 per year because the people running them aren't greedy lemmings trying to pay back VC.
And so the solution is… paying Anthropic $200/mo…
How are you completely disregarding the data integrity and privacy aspects of rolling your own tools?
I vibecode an app that only I use and store data locally. That means my data never leaves my device, I never have to share my email with anyone, never have to enter my credit card info anywhere
You buy SaaS and you have to then login, share credit card info, and have your data stored in the cloud somewhere with godknows what security practices
That’s worth more than the cost of any tokens
The "Your" shouldn't have been stripped from the title IMHO.
Depending on how you read the owner of an app subscription (the provider or the subscriber) maybe 'My' is better.
I believe it's a play on "Your margin is my opportunity".
Yeah. Part of why this is possible is simply that there are tons of subscription apps out there that were never really justified in requiring a recurring payment and are actually fairly trivial.
It used to be that you offer subscriptions only if there are ongoing costs, and a one-time payment if not (utilities, local, etc). SaaS kinda ruined that.
I'd welcome a boom in DIY vibe-coded utilities for personal use.
What I fear is a pollution of the open source space with tons of tailored apps that have a lot of overlap, but none of them get meaningful contributions because the maintainer will most likely respond with wontfix to almost everything (if they respond at all).
I build in the open, but what I build is just for me. If someone wants to fork it and modify it, they can go ahead - pretty much all of my stuff is MIT licensed by default.
But I'm not going to start adding features to my bespoke utility to fix someone else's problem.
Which is exactly why whenever I have an idea I just tinked with ClaudeCode for an hour or so until I have exactly what I need. It takes less time than trying to compare 10 similar products, none of which have the exact specifications or features that I need.
Subscription apps often have to target a wide userbase. However, most users only need a small subset of the entire feature set, and would be better served by a tailored version. This means that vibecoded apps can get away with being much less complex (specific featureset, no login etc), while still being more useful.
I have also created tools with LLMs that are exactly tailored to what I need, and still much more polished than what I could do without LLMs. Will have to think about if there is anything else I can do this with.
I had to download file attachments from a specific niche web forum that's behind login. I could've went looking for a browser plugin or a 3rd party tool to do it.
Once again, it took me about an hour while watching my shows to get a custom one made.
The first version operated by me downloading the pages one by one to a directory, the Python app parsed the html, downloaded the files and renamed according to thread name.
After a few iterations the tool just grabs a cookies.txt file exported from Firefox and can take any thread URL, browse through it, skipping existing files and determining if everything is already downloaded
I could easily have it just watch a set of threads for new content and download automatically, but the current system is fine =)
> I’m still skeptical of vibecoding in general. As I mentioned above, I would not trust my vibecoding enough to make these into products.
That’s the whole point - there’s no need for it to be a product when you can do it yourself, and it’s the death knell of products like this.
Also you dont need the architecture to scale to millions of users. So it can be a bit more rough and OK to be inefficient.
And you don't need to build in user management, Stripe integration for payments, build a landing page, advertise it or anything.
You can just use it and be content.
I made the same realization two weeks ago. Posted about it here, where I rebuilt bare bones todoist with a habit tracker, goal setting and more within a few vibe coding sessions: https://news.ycombinator.com/item?id=46633092
I think that many existing apps with huge userbases will gradually lose users as the models become better and better. Their biggest advantage is that people don't like change, and thus having to e.g. export data from some tools etc. seems to be a hassle not worth $5 a month.
But as the models get better and the quality of the output will match the quality of the established SaaS but tailor the whole thing to a single user with the ability to make any change they can imagine within minutes, and perhaps deploy to Hetzner and whatnot where they could host all of those apps for a single $5 instead, the exodus will accelerate.
On the other hand, new products will have a much harder time to gather a big userbase. Whenever I'll see a launch of a SaaS asking for $$$, the first question I'll ask myself will be how long it will take LLM to recreate it. And for most cases, I imagine that the time it will take to get 80% of what they have is a few vibe coding sessions (as most newcomers will probably have used LLM themselves to code it up).
> But as the models get better and the quality of the output will match the quality of the established SaaS but tailor the whole thing to a single user with the ability to make any change they can imagine within minutes, and perhaps deploy to Hetzner and whatnot where they could host all of those apps for a single $5 instead, the exodus will accelerate.
I do think you're vastly overestimating people's ability to write software, even with LLMs, and use it in production. The average computer user does not even use a computer as their primary computing device, they use a phone. The barrier to going from idea to phone app on iPhone or Android is relatively high.
Todo list apps, habit trackers, and the like are almost a special snowflake breed. Almost everyone has some different cross-section of needs they care about, and no app is perfect for each individual. So it's natural to say "is there something that matches what I want?" and then reach for tools to make that. The world is your oyster for todo list apps. Of course, the real issue comes from data sovereignty, trust, quality, things like that. When Apple launches a new device or a new iOS feature people want, you get to see which apps will actually implement the new features or which stagnate. They're a natural avenue for vibe coding since they're so particular.
Depends on how you define "production".
People in general would recoil in horror if they knew how many essential operations are backed by a mess of Excel sheets with formulas and VBA nobody understands anymore.
All it needs is the maker mindset of being just lazy enough to be bothered by a repetitive task and the courage (and permission) to use an Agentic LLM to figure out a fix for the issue.
It will be interesting to see if Apple/Android provide a platform for vibe-apps.
It's somewhat like the Shortcuts system on steroids.
> It will be interesting to see if Apple/Android provide a platform for vibe-apps.
It would be interesting, particularly for Apple, as this would cannibalize fees charged on the App Store. I imagine they could charge for use of the vibe-coding platform, but Apple hasn't been great at figuring out LLMs.
It would be cool if 3rd partly app platform could provide this functionality, but as I noted in another comment, I cannot even install my own vibe-coded apps to my own iPhone. (Without the 100 USD a year developer tax.) So I'm not sure how the architecture would work on iOS.
This is the same for me and I've not written code for years since I was a kid in school.
I vibe coded a webapp that I was paying yearly for and the version I made does everything I wish the app I paid for did as it's 100% personalised to me.
I've been thinking for awhile that this is going to be the future and I'm already starting to think of more things I will create.
can you share which paid webapp you were using and maybe a link to your version's github?
The App was TeuxDeux.
I haven't published it publicly yet, as i use it personally and it's a little flakey still, but will look to do in the future once I finish adding all the features.
My version is much more feature rich than TeuxDeux and I made it for Free in Google AI studio over the past two months, between other things. I'd just type a prompt and then go do some other stuff, it has taken quite a lot of revision but I haven't written a single line of code and i've been using it daily to manage my tasks since the start of December.
I see the sceptical comments, but no one says this "vibe-coded" projects/apps/tools will be ready for your customers. It basically scratches the itch for the given users/company/whatever. Also, it doesn't have to be fast, stable or handle 1_000_000 concurrent users. You don't have to worry about that.
Not everything has to be a SaaS, but I don't think all SaaS apps can be vibe-coded to a weekend project.
If it is solving my issues and problems, why do preaching about the merits of a proper product or paying. I'll pay for what I see value in, and vibe-code where I don't see the benefit of paying.
Maybe I miserably fail and get back to paying to product. It's all good, I take that responsibility while I start my vibe-coding session.
Turning a paid sub like WisprFlow into your own weekend build (Jabber) is a great move, but you can take it further by finding open-source alternatives that already implement the features you're replicating. For dictation and speech-to-text like what WisprFlow does, there's Handy[0], a free, open-source, offline speech-to-text app that runs locally with Whisper models.
Once you identify something like Handy, instruct Claude to study how that OSS project actually builds the feature and adapt the logic to your stack. AI is really good at finding the "seams" (those connection points where a feature ties into the tech stack) and understanding the full implementation.
The trick is knowing precisely where the feature lives in the code (files, functions, modules), because AIs often miss scattered pieces and don't capture everything otherwise. That's what I'm working on at opensource.builders[1]: turning OSS repos into a modular cookbook of features you can remix across stacks, with structured "skills" that point to the exact details so the porting works reliably.
So basically steal code from OSS, oh no I meant "get inspired by" OSS code without actually contributing anything. This is just gross as a developer imo.
We're big believers in personal software: even with OSS alts existing, building your own in your stack is valuable. Before AI, you'd read code from other apps, understand the implementation, and reimplement it yourself. This just makes that easier and faster.
My projects (Openship/Openfront[0]) are the first on the chopping block. We're creating modular OSS alts for every vertical (barbershops, hotels, etc.) for folks to take, remix, adapt, or fork into their tools. Chances are your AI model is already trained on similar OSS and building from it anyway. We make finding the exact code reliable. Check out our ethos to learn more [1].
I get that this is tempting but it just means you'll slowly get dependent on things that will eventually break in ways you will have no capacity to fix. And disaster recovery is most certainly a manual task.
> you'll slowly get dependent on things that will eventually break in ways you will have no capacity to fix
If the commercial provider charging you $10 a month breaks it, you also have no capacity to fix it.
Your options are: send them an email, or unsubscribe and use something else.
Right but most of the time, in my experience they keep the lights on.
Keeping the lights on is fine.
But if they remove a feature I rely on, I can't put it back.
If they add a feature I hate, I can't remove it.
If they jack the price up, I have no real solution to this.
If they move features I rely on from the standard tier to the 5x more expensive pro tier, I have no real solution to this.
Why, yes, this is an echo of the old argument for open source software.
Why wouldn't I be able to fix these things? If I managed to build a thing from scratch (with Opus 4.5), I don't see why I wouldn't be able to fix it and maintain it in the future (maybe with Opus 4.7 or even better future models?).
Why would they "eventually break"?
In what situation would a simple script or helper app just suddenly rot away and stop working?
Of course it's POSSIBLE to vibe together a massive monstrosity of an everything-app, but that's not what the author is doing here (nor me).
If you build enough things, you will also gain the experience to fix those things
I'm doing something very similar (creating my own apps for personal use), but I'm creating iOS apps primarily.
Here's what bugs me: I cannot permanently install my apps to my iPhone because of Apple's walled garden. I need to reinstall every 7 days and constantly re-confirm that I am a "Trusted" developer.
I know I can pay Apple 100 USD a year for a developer account, but I bought this phone outright 7 years ago, I own it. (Obviously, I clearly don't in this case.) /rant
Are you using features that can't be replicated with a PWA?
I actually created a PWA first, but it was just even more rough around the edges than the vibe-coded iOS app.
I wanted something that felt like an app, so would use iOS design elements, have widgets, use on-device storage (for offline use), etc. Apple, very intentionally I believe, makes a lot of these things harder than they need to be.
It’s easy to overlook what I think is the real value of these “home-built” tools.
We can now produce products and apps that are tailored to our own preferred ways of working.
Regardless of the cost of generating them (which can be as low as $20 per month for a ChatGPT Plus subscription) or the effort involved (sometimes less than an hour of “vibe coding”), we’ve reached a point where the resulting product can be significantly more valuable than the existing product, service, or subscription it replaces.
This reminds me so much of Maggie Appleton blog post on "Home-Cooked Software and Barefoot Developers" [1]
We have so many people who are so excellent and fast and developing nowadays that we can even afford the time to build things for our community, friends and even just for ourselves.
It has probably always been like this, but I am just personally observing a higher-degree of people doing and talking about it. Even just the small-web/neocities bobble points into this.
This hits close to home. I've been building tools for bookkeepers and accountants as a side project, and the calculus you're describing - where a subscription becomes a weekend obligation - is exactly why I've tried to keep things genuinely useful rather than sticky.
The cynical approach would be to make the product hard to leave. But that just means you've built a trap, not something people actually want. Eventually they escape and hate you for it.
The test I use: would people recommend this to colleagues even if there's no referral incentive? If the answer is no, I'm probably building something people tolerate rather than something they value.
I doubt LLM-generated software is going to replace more traditional software any time soon, especially when accuracy is pretty important (such as accounting). One thing I learned from years as a PM in a very data-centric organization is understanding data, how it is generated/stored/cut/etc. is very important to getting accurate results.
Where I could see some really interesting results is the marriage of the two. For example, you have a solid data structure that an LLM can generate infinite custom views from.
And I have first hand knowledge of well-known companies building their own tooling because the SaaS offerings have a bad price/feature ratio.
i think the same, i think backend where data is more prominent is not going anywhere soon. llms produce very bad data structures.
but from good apis, good data, good interface they can generate quite nice frontends.
i guess, frontend as job is going to have a hard time.
also, writing code is not cognitive load, its always reading code. and llms just increase that. so i mostly try to avoid using them.
but i do like researching with them. context free. like googles ai mode, etc. not from my code editor cause then they get biased and suggest stupid sh8t all the time.
You can pivot your knowledge into building bespoke tools for the same people, just a LOT faster.
The recommendation thing is a nice benchmark, but if you're building hyper-specific tools - why would people recommend them to anyone? If you build a tool for an accountant that does some very niche thing only they're bothered by, why would they recommend to the analyst or receptionist in the company?
> Then just yesterday, a friend of mine was telling me how he got tired of paying for Typora and decided to vibecode his own Markdown editor
But typora is actually one time purchase and one of the rare apps that is priced well with good business model.
They have probably best RTL support and I wanted like your friend to write my own focused markdown editor with RTL support using clause and made some progress but realized that the time and cost of doing this is not worth it. I just paid typora a week ago for $15.
But I understand the point and I use Claude to hack together personal tools all the time.
I've been evangelizing vibe coding, because we are wielding something much more powerful now than even ~3 months prior (Nov was the turning point).
Now that Prometheus (the myth, not the o11y tool) has dropped these LLMs on us, I've been using this thought experiment to consider the multi-layered implications:
In a world where everyone can cook, why would anybody buy prepared food?
This is where claude code is so good because all the stuff AI is bad at - security, auth, storage - are not a problem if its just you using it locally
Has anyone tried, on a large scale, designing the architecture of an application or writing behavioral tests (which effectively is designing the app through testing) and given the LLM the task of writing the implementation details?
Does that work better for maintainability than letting it decide on its own what the architecture should look like?
If so, what is your setup/workflow?
Loved the article, thanks for sharing. I’m curious if you’d share your setup. I haven’t made any macOS apps before, primarily because I never wanted to really learn XCode and obj-c. I like swift but still prefer simpler editors like Zed/VSC vs. What XCode offers.. so when you’re building these are you doing it in XCode or in another tool like Claude Code/codex/gemini CLI?
cheers
Not OP, but I use Xcode with Claude Pro and it is going fairly well. I also am creating my own personal-use apps instead of paying for monthly subscriptions. I know a bit of Swift, and had been trying to learn it while also using LLMs. At this point, I've decided to also not make these real projects and just vibe-code exactly what I want.
I did this recently at the company I work at. Someone suggested GitBook, I 'Vibe coded' an internal docs website in under and hour. Does what we need, looks good.
Unless the app has a large community/network and is just a SaaS with some offering, it'll be very easy to replace it.
It's interesting, because a few years ago I would have put this strictly under the "not invented here" fallacy, where we'd now be stuck maintaining another project for the foreseeable future. I used to press pretty hard to avoid it.
Now I wonder if the maintenance cost for this type of internal system has gone down to a level where that is no longer an issue.
I can see it going both ways.
If knowledge work continues to roughly look the way it looks then I think maintenance is going to be an issue. Both in terms of keeping the spaghetti together but also in terms of all the bad design decisions you get from everyone bolting on their ideas.
If however knowledge work becomes just talking to LLMs and occasionally interacting with an on the fly generated UI then maintenance becomes a non issue
We're currently paying through the nose for a shitty intranet solution that could've been just hugo with a markdown editor.
Or a vibe-coded simple website.
But "designed and implemented company-wide intranet" looks good in someone's CV so here we are.
The prevalence of this "personal vibecoded app" spirit makes me start to wonder if an "App" is the right level of abstraction for packaging capabilities. Perhaps we need something more "granular".
Just sucks that instead of buying a piece of software now we're going to send $100/mo to an AI giant so we can build our own crappier bespoke apps.
And one day your day job will be someone else weekend project
Only assuming a sufficient number of your customers is happy and capable to vibe-code a product for their set of the requirements that your product solves.
I should vibecode my own Onshape.
Some apps just do not make any sense for monetization and are now raced to zero. If it's not already open sourced then someone will vibe-code an implementation if none exist.
Vibe-coding accelerates the destruction of basic (closed-source) apps charging a subscription for features that offer little to no value whatsoever.
The SaaS business model took things too far anyway. Everything is a subscription and it gets tiring quickly. I am glad that LLMs can replace crappy SaaS with crappy code now.
I replaced a whole bunch of these with one shot prompts for shits and giggles.
I think there's still an underestimated burden to vibe-coding an app for a non-software engineer. I'm not recommending my parents vibe-code apps to solve problems, so I think the market is smaller.
But Roberto's use-case is definitely more sane than most.
I'm currently assisting three very non-programmer people in-house who did just this.
Their problem is solved, now it's up to me to update the internal guidelines and agent instructions so that the code is at least semi-decent.
None of these are going to "production", they all live on local company controlled laptops and only one of them might access an external API automatically later this spring.
But each of them takes hours of manual work and does it in minutes.
With the current tech, I agree this will still be pretty niche. I'm vibe-coding my own iOS apps, and it still needs a decent understanding of the tech and a willingness to put up with a lot of rough edges.
However, with a proper framework (e.g., a very opinionated design system, the ability to choose from some pre-designed structures/flows, etc.) I could very much see ad hoc creation of software becoming more widespread.
Thats not cheaper than paying a subscription.
In fact this is at least 3x-10x more expensive.
And this is comparing to being subscribed many years in a row.
With SaaS you can unsub and sub only when you need it again.
With your side project - a weekend of your life is invested and you will never get it back.
This is the worst use of your time if you measure it in $.
If you make it for fun - sure. In all other terms it is a complete loss.
$20 claude code subscription for a month can replace the $15 + $10 for each month. How is that 3x more espensive? The user just saved $280 per year, on just two subscriptions alone.
Hardly doubt that this was the 'most waste of ones time'. For one, it's not like most of us can decide to "work" for 3-5 hours on a Saturday and get any money. I play games on my pc while claude codes for me. I alt tab each few minutes and see if it needs any input. Then I can (not that I do it), read and perhaps learn from the code.
Hey, CC has a way to trigger a notification chime on completion.
(How? Idk, I just asked it to guide me through the short hook process)
Afaik it's built in now.
I get a notification on macOS with the title of the context.
Yup. thats more expensive because each hour of your time is at least 50USD
And each hour on weekend that you would have spent with your family etc is probably 500-1000 usd at least, so yeah, it is much cheaper to pay 15 usd for SaaS
Software is a manifestation of someone’s knowledge of and experience in and ideas about how a thing should work. We learn from the software we use, we benefit from everyone else’s ideas, we benefit from the hundreds and thousands of hours other people put into understanding a problem to design a solution. My workflow is better because of the incremental improvements made by developer after developer year after year. Would we have Claude Code if our foredevelopers hadn’t spent thousands of hours deep in thought, obsessing over every last detail?
Building all the software you use yourself, whether by hand or by vibe coding, cuts you off from the world.
I have no philosophical objection to vibe-coding apps for yourself, but personally, I wouldn’t be 1/10th of the engineer I am if I wasn’t constantly exposed to the work of others.
For some, this trend worries software engineers — who needs software if they can vibe code it themselves? — but I am much more optimistic. I think people will start valuing good software a lot more. Claude code can deliver the first 90%, but we all know it is the last 90% that differentiates.
I like to say "my code is 200% vibe-coded; the tricky bit is figuring out which 100% to keep".
Decisions matter, both technical and product ones. LLMs don't make as good technical or product decisions as I would, and the way I work with them tries to maximize my strengths and the LLM's strengths. I don't know if I succeed, but it's better than "make me an app like X" as a prompt.
Your entire post is self selection bias and survivorship bias.
SWE field is one of the most cognitive dissonant social groups; cries foul at the slightest whiff their free speech and agency is being put upon; seeks to reduce blockers to their productivity, fewer PMs! Less management!
Now complains about users using their machines without having to block on an SWE.
Insert that quote about how someone will not see the obvious if their paycheck relies on them ignoring the obvious.
Here come LLMs and all they can accomplish with a few arithmetical rules instead of the arbitrary semantics of an SWE; watch as SWEs block social evolution away from disrupting software engineers.
As an example; "protected memory", among many other individual software problems, is an access control problem mired in old semantics relative to OS monoliths.
Didn’t see you all halting as you decimated travel agent jobs, retail jobs, etc etc. Technology advancement must now stand still after centuries of evolution? The self selection bias is as obvious as Trump's.
Making your own software is a good way to escape enshittification and influence.
I switched from Spotify to buying MP3s and using my own audio client, because I'm fed up of a company telling me which music I should listen to every single time I open the app. It costs more, but I own the music and I escape the constant redesigns, price increases and influential behaviour.
Most apps are very simple and there isn't too much to learn, especially if you're building it to scale to a userbase of yourself. I can't see the need for a ton of CRUD apps which demand subscription fees personally. If you build them yourself, you get to keep your own data, build it out the way you want it, keep it that way, and use computers as a person using a tool as opposed to a customer buying a product.
First of all, I’m skeptical about these being free. Time isn’t free, and the tokens to make these projects certainly weren’t free.
Second of all, all of these SaaS apps that don’t actually have a need for recurring charge probably should be paid one time. I don’t use Loom — I use CleanShot X and it was a one-time $30 payment and has a lot of great features I benefit from. I can’t reimplement it in $30 of tokens or $30 of my time.
But for an app whose use case doesn’t change and is recurring for no reason? Yeah there’s probably not much value in recurring payments outside of wanting to support the developer. I pay a lot of indie devs out of the goodness of my heart, and I’ll continue to do that.
But the value for “SaaS apps” without clear monthly costs should have always been under scrutiny.
Most apps rationale for subscriptions is "Ongoing development" without an option like jetbrains etc. to fall back to a perpetual license. In practice, regardless of whether an app needs ongoing development or not, this is the best way to try to guarantee continuous income and make a living off of a project I guess.
Perhaps LLM's will force developers/companies to change their stance and to stop users from recreating what they have already created, just buy an at-a-time snapshot of their app for a one-time-fee? Probably not but one can hope.
It is not irrational on the part of the developer -- I've definitely felt this too. The problem comes from the fact that practically everyone has subscription fatigue these days, and each of us probably has only a few pieces of software we truly care about enough to want to support them out of the goodness of our hearts.
But everyone wants us to pay $10/mo. It just isn't sustainable from a consumer perspective.
> But everyone wants us to pay $10/mo. It just isn't sustainable from a consumer perspective.
And so few actually deliver $10/mo worth of value. If 1password and Fastmail - the two most important services that control my digital life - are each $60/year, that's the standard of value other SaaS companies have to beat and very few do. The ones that do are like NextDNS where they cost $20-30 per year because the people running them aren't greedy lemmings trying to pay back VC.
And so the solution is… paying Anthropic $200/mo…
How are you completely disregarding the data integrity and privacy aspects of rolling your own tools?
I vibecode an app that only I use and store data locally. That means my data never leaves my device, I never have to share my email with anyone, never have to enter my credit card info anywhere
You buy SaaS and you have to then login, share credit card info, and have your data stored in the cloud somewhere with godknows what security practices
That’s worth more than the cost of any tokens
The "Your" shouldn't have been stripped from the title IMHO.
Depending on how you read the owner of an app subscription (the provider or the subscriber) maybe 'My' is better.
I believe it's a play on "Your margin is my opportunity".
Yeah. Part of why this is possible is simply that there are tons of subscription apps out there that were never really justified in requiring a recurring payment and are actually fairly trivial.
It used to be that you offer subscriptions only if there are ongoing costs, and a one-time payment if not (utilities, local, etc). SaaS kinda ruined that.
I'd welcome a boom in DIY vibe-coded utilities for personal use.
What I fear is a pollution of the open source space with tons of tailored apps that have a lot of overlap, but none of them get meaningful contributions because the maintainer will most likely respond with wontfix to almost everything (if they respond at all).
I build in the open, but what I build is just for me. If someone wants to fork it and modify it, they can go ahead - pretty much all of my stuff is MIT licensed by default.
But I'm not going to start adding features to my bespoke utility to fix someone else's problem.
Which is exactly why whenever I have an idea I just tinked with ClaudeCode for an hour or so until I have exactly what I need. It takes less time than trying to compare 10 similar products, none of which have the exact specifications or features that I need.
List of projects mentioned before: https://news.ycombinator.com/item?id=46716805
This is a fun article and approach.
Subscription apps often have to target a wide userbase. However, most users only need a small subset of the entire feature set, and would be better served by a tailored version. This means that vibecoded apps can get away with being much less complex (specific featureset, no login etc), while still being more useful.
I have also created tools with LLMs that are exactly tailored to what I need, and still much more polished than what I could do without LLMs. Will have to think about if there is anything else I can do this with.
I had to download file attachments from a specific niche web forum that's behind login. I could've went looking for a browser plugin or a 3rd party tool to do it.
Once again, it took me about an hour while watching my shows to get a custom one made.
The first version operated by me downloading the pages one by one to a directory, the Python app parsed the html, downloaded the files and renamed according to thread name.
After a few iterations the tool just grabs a cookies.txt file exported from Firefox and can take any thread URL, browse through it, skipping existing files and determining if everything is already downloaded
I could easily have it just watch a set of threads for new content and download automatically, but the current system is fine =)
> I’m still skeptical of vibecoding in general. As I mentioned above, I would not trust my vibecoding enough to make these into products.
That’s the whole point - there’s no need for it to be a product when you can do it yourself, and it’s the death knell of products like this.
Also you dont need the architecture to scale to millions of users. So it can be a bit more rough and OK to be inefficient.
And you don't need to build in user management, Stripe integration for payments, build a landing page, advertise it or anything.
You can just use it and be content.
I made the same realization two weeks ago. Posted about it here, where I rebuilt bare bones todoist with a habit tracker, goal setting and more within a few vibe coding sessions: https://news.ycombinator.com/item?id=46633092
I think that many existing apps with huge userbases will gradually lose users as the models become better and better. Their biggest advantage is that people don't like change, and thus having to e.g. export data from some tools etc. seems to be a hassle not worth $5 a month. But as the models get better and the quality of the output will match the quality of the established SaaS but tailor the whole thing to a single user with the ability to make any change they can imagine within minutes, and perhaps deploy to Hetzner and whatnot where they could host all of those apps for a single $5 instead, the exodus will accelerate.
On the other hand, new products will have a much harder time to gather a big userbase. Whenever I'll see a launch of a SaaS asking for $$$, the first question I'll ask myself will be how long it will take LLM to recreate it. And for most cases, I imagine that the time it will take to get 80% of what they have is a few vibe coding sessions (as most newcomers will probably have used LLM themselves to code it up).
> But as the models get better and the quality of the output will match the quality of the established SaaS but tailor the whole thing to a single user with the ability to make any change they can imagine within minutes, and perhaps deploy to Hetzner and whatnot where they could host all of those apps for a single $5 instead, the exodus will accelerate.
I do think you're vastly overestimating people's ability to write software, even with LLMs, and use it in production. The average computer user does not even use a computer as their primary computing device, they use a phone. The barrier to going from idea to phone app on iPhone or Android is relatively high.
Todo list apps, habit trackers, and the like are almost a special snowflake breed. Almost everyone has some different cross-section of needs they care about, and no app is perfect for each individual. So it's natural to say "is there something that matches what I want?" and then reach for tools to make that. The world is your oyster for todo list apps. Of course, the real issue comes from data sovereignty, trust, quality, things like that. When Apple launches a new device or a new iOS feature people want, you get to see which apps will actually implement the new features or which stagnate. They're a natural avenue for vibe coding since they're so particular.
Depends on how you define "production".
People in general would recoil in horror if they knew how many essential operations are backed by a mess of Excel sheets with formulas and VBA nobody understands anymore.
All it needs is the maker mindset of being just lazy enough to be bothered by a repetitive task and the courage (and permission) to use an Agentic LLM to figure out a fix for the issue.
It will be interesting to see if Apple/Android provide a platform for vibe-apps.
It's somewhat like the Shortcuts system on steroids.
> It will be interesting to see if Apple/Android provide a platform for vibe-apps.
It would be interesting, particularly for Apple, as this would cannibalize fees charged on the App Store. I imagine they could charge for use of the vibe-coding platform, but Apple hasn't been great at figuring out LLMs.
It would be cool if 3rd partly app platform could provide this functionality, but as I noted in another comment, I cannot even install my own vibe-coded apps to my own iPhone. (Without the 100 USD a year developer tax.) So I'm not sure how the architecture would work on iOS.
This is the same for me and I've not written code for years since I was a kid in school.
I vibe coded a webapp that I was paying yearly for and the version I made does everything I wish the app I paid for did as it's 100% personalised to me.
I've been thinking for awhile that this is going to be the future and I'm already starting to think of more things I will create.
can you share which paid webapp you were using and maybe a link to your version's github?
The App was TeuxDeux.
I haven't published it publicly yet, as i use it personally and it's a little flakey still, but will look to do in the future once I finish adding all the features.
My version is much more feature rich than TeuxDeux and I made it for Free in Google AI studio over the past two months, between other things. I'd just type a prompt and then go do some other stuff, it has taken quite a lot of revision but I haven't written a single line of code and i've been using it daily to manage my tasks since the start of December.
I see the sceptical comments, but no one says this "vibe-coded" projects/apps/tools will be ready for your customers. It basically scratches the itch for the given users/company/whatever. Also, it doesn't have to be fast, stable or handle 1_000_000 concurrent users. You don't have to worry about that.
Not everything has to be a SaaS, but I don't think all SaaS apps can be vibe-coded to a weekend project.
If it is solving my issues and problems, why do preaching about the merits of a proper product or paying. I'll pay for what I see value in, and vibe-code where I don't see the benefit of paying.
Maybe I miserably fail and get back to paying to product. It's all good, I take that responsibility while I start my vibe-coding session.
Turning a paid sub like WisprFlow into your own weekend build (Jabber) is a great move, but you can take it further by finding open-source alternatives that already implement the features you're replicating. For dictation and speech-to-text like what WisprFlow does, there's Handy[0], a free, open-source, offline speech-to-text app that runs locally with Whisper models.
Once you identify something like Handy, instruct Claude to study how that OSS project actually builds the feature and adapt the logic to your stack. AI is really good at finding the "seams" (those connection points where a feature ties into the tech stack) and understanding the full implementation.
The trick is knowing precisely where the feature lives in the code (files, functions, modules), because AIs often miss scattered pieces and don't capture everything otherwise. That's what I'm working on at opensource.builders[1]: turning OSS repos into a modular cookbook of features you can remix across stacks, with structured "skills" that point to the exact details so the porting works reliably.
[0] https://github.com/cjpais/Handy
[1] https://github.com/junaid33/opensource.builders
So basically steal code from OSS, oh no I meant "get inspired by" OSS code without actually contributing anything. This is just gross as a developer imo.
We're big believers in personal software: even with OSS alts existing, building your own in your stack is valuable. Before AI, you'd read code from other apps, understand the implementation, and reimplement it yourself. This just makes that easier and faster.
My projects (Openship/Openfront[0]) are the first on the chopping block. We're creating modular OSS alts for every vertical (barbershops, hotels, etc.) for folks to take, remix, adapt, or fork into their tools. Chances are your AI model is already trained on similar OSS and building from it anyway. We make finding the exact code reliable. Check out our ethos to learn more [1].
[0] https://openship.org
[1] https://opensource.builders/ethos
I get that this is tempting but it just means you'll slowly get dependent on things that will eventually break in ways you will have no capacity to fix. And disaster recovery is most certainly a manual task.
> you'll slowly get dependent on things that will eventually break in ways you will have no capacity to fix
If the commercial provider charging you $10 a month breaks it, you also have no capacity to fix it.
Your options are: send them an email, or unsubscribe and use something else.
Right but most of the time, in my experience they keep the lights on.
Keeping the lights on is fine.
But if they remove a feature I rely on, I can't put it back.
If they add a feature I hate, I can't remove it.
If they jack the price up, I have no real solution to this.
If they move features I rely on from the standard tier to the 5x more expensive pro tier, I have no real solution to this.
Why, yes, this is an echo of the old argument for open source software.
Why wouldn't I be able to fix these things? If I managed to build a thing from scratch (with Opus 4.5), I don't see why I wouldn't be able to fix it and maintain it in the future (maybe with Opus 4.7 or even better future models?).
Why would they "eventually break"?
In what situation would a simple script or helper app just suddenly rot away and stop working?
Of course it's POSSIBLE to vibe together a massive monstrosity of an everything-app, but that's not what the author is doing here (nor me).
If you build enough things, you will also gain the experience to fix those things
I'm doing something very similar (creating my own apps for personal use), but I'm creating iOS apps primarily.
Here's what bugs me: I cannot permanently install my apps to my iPhone because of Apple's walled garden. I need to reinstall every 7 days and constantly re-confirm that I am a "Trusted" developer.
I know I can pay Apple 100 USD a year for a developer account, but I bought this phone outright 7 years ago, I own it. (Obviously, I clearly don't in this case.) /rant
Are you using features that can't be replicated with a PWA?
I actually created a PWA first, but it was just even more rough around the edges than the vibe-coded iOS app.
I wanted something that felt like an app, so would use iOS design elements, have widgets, use on-device storage (for offline use), etc. Apple, very intentionally I believe, makes a lot of these things harder than they need to be.
It’s easy to overlook what I think is the real value of these “home-built” tools.
We can now produce products and apps that are tailored to our own preferred ways of working.
Regardless of the cost of generating them (which can be as low as $20 per month for a ChatGPT Plus subscription) or the effort involved (sometimes less than an hour of “vibe coding”), we’ve reached a point where the resulting product can be significantly more valuable than the existing product, service, or subscription it replaces.
This reminds me so much of Maggie Appleton blog post on "Home-Cooked Software and Barefoot Developers" [1]
We have so many people who are so excellent and fast and developing nowadays that we can even afford the time to build things for our community, friends and even just for ourselves.
It has probably always been like this, but I am just personally observing a higher-degree of people doing and talking about it. Even just the small-web/neocities bobble points into this.
[1] https://maggieappleton.com/home-cooked-software
This hits close to home. I've been building tools for bookkeepers and accountants as a side project, and the calculus you're describing - where a subscription becomes a weekend obligation - is exactly why I've tried to keep things genuinely useful rather than sticky.
The cynical approach would be to make the product hard to leave. But that just means you've built a trap, not something people actually want. Eventually they escape and hate you for it.
The test I use: would people recommend this to colleagues even if there's no referral incentive? If the answer is no, I'm probably building something people tolerate rather than something they value.
I doubt LLM-generated software is going to replace more traditional software any time soon, especially when accuracy is pretty important (such as accounting). One thing I learned from years as a PM in a very data-centric organization is understanding data, how it is generated/stored/cut/etc. is very important to getting accurate results.
Where I could see some really interesting results is the marriage of the two. For example, you have a solid data structure that an LLM can generate infinite custom views from.
https://www.databricks.com is doing this already with data as well as multiple other companies
And I have first hand knowledge of well-known companies building their own tooling because the SaaS offerings have a bad price/feature ratio.
i think the same, i think backend where data is more prominent is not going anywhere soon. llms produce very bad data structures.
but from good apis, good data, good interface they can generate quite nice frontends.
i guess, frontend as job is going to have a hard time.
also, writing code is not cognitive load, its always reading code. and llms just increase that. so i mostly try to avoid using them.
but i do like researching with them. context free. like googles ai mode, etc. not from my code editor cause then they get biased and suggest stupid sh8t all the time.
You can pivot your knowledge into building bespoke tools for the same people, just a LOT faster.
The recommendation thing is a nice benchmark, but if you're building hyper-specific tools - why would people recommend them to anyone? If you build a tool for an accountant that does some very niche thing only they're bothered by, why would they recommend to the analyst or receptionist in the company?
> Then just yesterday, a friend of mine was telling me how he got tired of paying for Typora and decided to vibecode his own Markdown editor
But typora is actually one time purchase and one of the rare apps that is priced well with good business model.
They have probably best RTL support and I wanted like your friend to write my own focused markdown editor with RTL support using clause and made some progress but realized that the time and cost of doing this is not worth it. I just paid typora a week ago for $15.
But I understand the point and I use Claude to hack together personal tools all the time.
I've been evangelizing vibe coding, because we are wielding something much more powerful now than even ~3 months prior (Nov was the turning point).
Now that Prometheus (the myth, not the o11y tool) has dropped these LLMs on us, I've been using this thought experiment to consider the multi-layered implications:
In a world where everyone can cook, why would anybody buy prepared food?
This is where claude code is so good because all the stuff AI is bad at - security, auth, storage - are not a problem if its just you using it locally
Has anyone tried, on a large scale, designing the architecture of an application or writing behavioral tests (which effectively is designing the app through testing) and given the LLM the task of writing the implementation details?
Does that work better for maintainability than letting it decide on its own what the architecture should look like?
If so, what is your setup/workflow?
Loved the article, thanks for sharing. I’m curious if you’d share your setup. I haven’t made any macOS apps before, primarily because I never wanted to really learn XCode and obj-c. I like swift but still prefer simpler editors like Zed/VSC vs. What XCode offers.. so when you’re building these are you doing it in XCode or in another tool like Claude Code/codex/gemini CLI?
cheers
Not OP, but I use Xcode with Claude Pro and it is going fairly well. I also am creating my own personal-use apps instead of paying for monthly subscriptions. I know a bit of Swift, and had been trying to learn it while also using LLMs. At this point, I've decided to also not make these real projects and just vibe-code exactly what I want.
I did this recently at the company I work at. Someone suggested GitBook, I 'Vibe coded' an internal docs website in under and hour. Does what we need, looks good. Unless the app has a large community/network and is just a SaaS with some offering, it'll be very easy to replace it.
It's interesting, because a few years ago I would have put this strictly under the "not invented here" fallacy, where we'd now be stuck maintaining another project for the foreseeable future. I used to press pretty hard to avoid it.
Now I wonder if the maintenance cost for this type of internal system has gone down to a level where that is no longer an issue.
I can see it going both ways. If knowledge work continues to roughly look the way it looks then I think maintenance is going to be an issue. Both in terms of keeping the spaghetti together but also in terms of all the bad design decisions you get from everyone bolting on their ideas. If however knowledge work becomes just talking to LLMs and occasionally interacting with an on the fly generated UI then maintenance becomes a non issue
We're currently paying through the nose for a shitty intranet solution that could've been just hugo with a markdown editor.
Or a vibe-coded simple website.
But "designed and implemented company-wide intranet" looks good in someone's CV so here we are.
The prevalence of this "personal vibecoded app" spirit makes me start to wonder if an "App" is the right level of abstraction for packaging capabilities. Perhaps we need something more "granular".
Just sucks that instead of buying a piece of software now we're going to send $100/mo to an AI giant so we can build our own crappier bespoke apps.
And one day your day job will be someone else weekend project
Only assuming a sufficient number of your customers is happy and capable to vibe-code a product for their set of the requirements that your product solves.
I should vibecode my own Onshape.
Some apps just do not make any sense for monetization and are now raced to zero. If it's not already open sourced then someone will vibe-code an implementation if none exist.
Vibe-coding accelerates the destruction of basic (closed-source) apps charging a subscription for features that offer little to no value whatsoever.
The SaaS business model took things too far anyway. Everything is a subscription and it gets tiring quickly. I am glad that LLMs can replace crappy SaaS with crappy code now.
I replaced a whole bunch of these with one shot prompts for shits and giggles.
I think there's still an underestimated burden to vibe-coding an app for a non-software engineer. I'm not recommending my parents vibe-code apps to solve problems, so I think the market is smaller.
But Roberto's use-case is definitely more sane than most.
I'm currently assisting three very non-programmer people in-house who did just this.
Their problem is solved, now it's up to me to update the internal guidelines and agent instructions so that the code is at least semi-decent.
None of these are going to "production", they all live on local company controlled laptops and only one of them might access an external API automatically later this spring.
But each of them takes hours of manual work and does it in minutes.
With the current tech, I agree this will still be pretty niche. I'm vibe-coding my own iOS apps, and it still needs a decent understanding of the tech and a willingness to put up with a lot of rough edges.
However, with a proper framework (e.g., a very opinionated design system, the ability to choose from some pre-designed structures/flows, etc.) I could very much see ad hoc creation of software becoming more widespread.
Thats not cheaper than paying a subscription. In fact this is at least 3x-10x more expensive.
And this is comparing to being subscribed many years in a row. With SaaS you can unsub and sub only when you need it again.
With your side project - a weekend of your life is invested and you will never get it back.
This is the worst use of your time if you measure it in $. If you make it for fun - sure. In all other terms it is a complete loss.
$20 claude code subscription for a month can replace the $15 + $10 for each month. How is that 3x more espensive? The user just saved $280 per year, on just two subscriptions alone.
Hardly doubt that this was the 'most waste of ones time'. For one, it's not like most of us can decide to "work" for 3-5 hours on a Saturday and get any money. I play games on my pc while claude codes for me. I alt tab each few minutes and see if it needs any input. Then I can (not that I do it), read and perhaps learn from the code.
Hey, CC has a way to trigger a notification chime on completion.
(How? Idk, I just asked it to guide me through the short hook process)
Afaik it's built in now.
I get a notification on macOS with the title of the context.
Yup. thats more expensive because each hour of your time is at least 50USD And each hour on weekend that you would have spent with your family etc is probably 500-1000 usd at least, so yeah, it is much cheaper to pay 15 usd for SaaS