Hacker News Reader

Comment by ulf-77723 | Hacker News Reader

Software is still eating the world, now even faster. I wonder how soon we will adapt to this new situation where software is vibe coded for anything and make use of this software without caution as expressed in the article.

For most people the main difference will be: Will it run and solve my problem? Soon we will see malware being put into vibe coded software - who will wants to check every commit for write-only software?

View on HN

I think in the future (in 10 years?) we are going to see a lot of disposable/throwaway software. I don’t know, imagine this: I need to buy tickets for a concert. I ask my AI agent that I want tickets. The agent creates code on the fly and uses it to purchase my tickets. The code could be simple curl command, or a full app with nice ui/ux. As a user I don’t need to see the code.

If I want to buy more tickets the same day, the ai agent will likely reuse the same code. But if i buy tickets again in one year, the agent will likely rebuild the code to adjust to the new API version the ticket company now offers. Seems wasteful but it’s more dynamic. Vendors only need to provide raw APIs and your agent can create the ui experience you want. In that regard nobody but the company that owns your agent can inject malware into the software you use. Some software will last more than others (e.g., the music player your agent provided won’t probably be rebuilt unless you want a new look and feel or extra functionality). I think we’ll adopt the “cattle, not pets” approach to software too.

This is also where I think we end up. If the behavior of the system is specified well enough, then the code itself is cheap and throwaway. Why have a static system that is brittle to external changes when you can just reconstruct the system on the fly?

Might be quite awhile before you can do this with large systems but we already see this on smaller contextual scales such as Claude Code itself

The specification for most systems _is the code_. English cannot describe business rules as succinctly as code, and most business rules end up being implied from a spec rather than directly specified, at least in my experience.

The thought of converting an app back into a spec document or list of feature requests seems crazy to me.

Why would it be? If you can describe an approximation of a system and regenerate it to be, let’s say, 98% accurate in 1% of the time that it would take to generate it by hand (and that’s being generous, it’s probably more like 0.1% in today’s day and age and that decimal is only moving left) aren’t there a giant set of use cases where the approximation of the system is totally fine? People will always bring up “but what about planes and cars and medicine and critical life or death systems”. Yeah sure, but a vast majority of the systems an end user interacts with every day do not have that level of risk tolerance

You are just validating the point that code is spec.

For your proposed system to work one must have a deterministic way of sending said spec to a system(Compiler?) and getting the same output everytime.

Input/Output is just one thing, software does a lot of 'side effect' kind of work, and has security implications. You don't leave such things to luck. Things either work or don't.

Absolutely let’s not do away with the determinism entirely. But we can decouple generation of the code from its deterministic behavior. If you are adequately able to identify the boundaries of the system and run deterministic tests to validate those boundaries that should be sufficient enough. It’s not like human written code was often treated with even that much scrutiny in the before times. I would validate human written code in the exact same way.

> If the behavior of the system is specified well enough

Then it becomes code: a precise symbolic representation of a process that can be unambiguously interpreted by a computer. If there is ambiguity, then that will be unsuitable for many systems.

The word “many” is carrying a lot of weight here. Given the probabilistic nature of AI I suspect that systems that are 98% correct will be just fine for all but the “this plane will crash” or “this person will get cancer” use cases. If the recreation of the system failed in that 2% by slightly annoying some end user, who gives a shit? If the stakes are low, and indeed they are for a large majority of software use cases, probabilistic approximation of everyone’s open source will do just fine.

If you’re worried about them achieving the 98%, worry no more, due to the probabilistic nature it will eventually converge on 9’s. Just keep sending the system through the probabilistic machine until it reaches your desired level of nines

>>If the behavior of the system is specified well enough, then the code itself is cheap and throwaway. Why have a static system that is brittle to external changes when you can just reconstruct the system on the fly?

You mean to say if the unit and functional tests cases are given the system must generate code for you? You might want to look at Prolog in that case.

>>Might be quite awhile before you can do this with large systems but we already see this on smaller contextual scales such as Claude Code itself

We have been able to do something like this reliably for like 50 years now.

Or, and hear me out here, you go to the existing site or app which sells concert tickets, press the purchase button, and then you have your tickets.

Like what are we even doing here...

I know people have done truly amazing things with AI lately, but I feel this in my bones. Almost every demo I see is like, uh, I don't need these extremely simple things in my life automated. I can just go to Delta and buy a plane ticket. I actually want to write my own email to my mom or wife. Of course a demo is just a demo, but also come on

I think it's a fallacy that if you make creating anything easier, more useful things will be created. In reality, you just end up with more useless things being created. Like with art, when it gets easier to create you don't end up with more good art. And with software - it's not like the quality of software has gone up as it's gotten easier to build, it's gotten much worse.

A related fallacy is that great things are easier to build when you can rapidly create stuff. That isn't really how great ideas are generated, it's not a slot machine where if you pull the lever 1000 times you generate a good idea and thus a successful piece of software can be made. This seems like a distinctly Silicon Valley, SFBA type mentality. Steve Jobs didn't invent the iPhone by creating 1000 different throwaway products to test the market. Etc etc.

> I think it's a fallacy that if you make creating anything easier, more useful things will be created. In reality, you just end up with more useless things being created.

Well, if you lower the competence bar required to do something, then more people of lower competence will do that thing.

It's easy to buy one plane ticket when a person has a specific plan -- to attend a meeting or a conference, or to match up with an airbnb timeslot or something.

It's harder to buy one plane ticket for the lowest cost amongst all the different ways that plane tickets can be bought, and harder yet to do so with a lack of specificity.

So, for instance: Maybe I don't have a firm plan. Maybe I'm very flexible.

Maybe all I want to do is say "Hey, bot. I want to go visit my friend in Florida sometime in the next couple of weeks and spend a few days there as inexpensively as I can. He's in Orlando. I can fly out of Detroit or Cleveland; all the same to me. If I drive to the airport myself, I'll need a place to keep my car at or near the airport. I also want to explore renting a car in Orlando. I pack light; personal bag only. Cattle class is OK, but I prefer a window seat. Present to me a list of the cheapest options, with itinerary."

That's all stuff that a human can sort out, but it takes time to manually fudge around dates and locations and deal with different systems and tabulate the results. And there's nuances that need covered, like parking at DTW is weird: It's all off-site, and it can be cheaper and better to rent a room for one night in a nearby hotel that includes long-term parking than to pay for parking by itself.

So the hypothetical bot does a bunch of API hits, applies its general knowledge of how things flow, and comes back with a list of verified-good options for me to review. And then I get to pick around that list, and ask questions, and mold it to best fit my ideal vision of an inexpensive trip to go spend time with a friend.

In English, and without ever dealing with any travel websites myself.

"Right. So I go to Detroit on Tuesday and check in at the hotel any time after noon, and take the free shuttle to the airport the next morning at around 0400 to the Evans terminal. Also, thanks for pointing out that this airport is like a ghost town until 0600 and I might want to bring a snack. Anyway, I get on the flight, land at Orlando, and they'll have a cheap car waiting for me at Avis. This will all cost me a total of $343, which sounds great. If that's all I need to know right now, then make it so. Pay for it and put it on my calendar."

(And yeah, this is a problem that I actually have from time to time. I'd love to have a bot that could just sort this stuff out with a few paragraphs.)

But who is really going to put together the infrastructure and harness to make all that work? My dad certainly isn't. My mother in law won't

What you describe will just end up a feature on Expedia. The highly technical builders of stuff that love to tinker vastly overestimate how much BS the general public will put up with

Indeed. I have zero desire to put such a thing together just for my own use.

I didn't address that concept at all above, but I think the notion of a million people each independently using the bot to write a million bespoke programs that each do the same things is...kind of a non-starter. It's something that can only happen in some weird reality where software isn't essentially free to copy, and where people are motivated neither by laziness, nor the size of their pocketbook.

If/when someone does put the work into getting it to happen, then I expect to find it on Github for people to lazily copy and use, or for them to make it available as a website or app for anybody to use (with even more laziness) -- and for them to monetize it.

My point is: such apps wouldn’t need to exist if agents can provide in the future the same functionality for a fraction of the cost. Sure if ticketmaster is here to stay forever and keep their app up to date, we can keep using it. But what about new products? Would companies decide to build a single fixed app that all the users have to use, instead of, well, not building it? Sure the functionality would still need to be provided by the company (e.g., like offered in the form of an api), so they keep getting profit.

It’s like we usually say: companies should focus on their core value. And typically the ui/ux is not the core value of companies.

> And typically the ui/ux is not the core value of companies

Huh? The user experience is basically ALL of the core product of a company.

If it's so easy for an AI to create ticket purchasing software that people can generate it themselves, then it's also true that the company can also use AI to generate that software for users who then don't need to generate it themselves. Obviously I think neither of these things are true or likely to happen.

> Huh? The user experience is basically ALL of the core product of a company.

Thats the case now, but I think it’s because there’s no other way around it nowadays. But if agents in the future provide a better or more natural ui/ux for many use cases, then companies core value will shift more into their inner core (which in software translates typically to the domain model)

> If it's so easy for an AI to create ticket purchasing software that people can generate it themselves, then it's also true that the company can also use AI to generate that software for users who then don't need to generate it themselves.

I think the generation of software per se will be transparent to the user. Users won’t think in terms of software created but wishes their agents make true.

So we burn the planet up to deploy individually craft UIs on demand? I mean, I've read your comment three times, and I just don't see it. If we end up in that future, we're doomed.

Why would I do that if the gateway to the internet becomes these LLM interfaces? How is it not easier to ask or type 'buy me tickets for Les Mis'? In the ideal world it will just figure it out, or I frustratingly have to interact with a slightly different website to purchase tickets for each separate event I want to see.

One of the benefits that I see is as much as I love tech and writing software, I really really do not want to interface with a vast majority of the internet that has been designed to show the maximum amount of ads in the given ad space.

The internet sucks now, anything that gets me away from having ads shoved in my face constantly and surrounded by uncertainty that you could always be talking to a bot.

I'm sympathetic to this view too, but I don't think the solution is to have LLM's generate bespoke code to do it. We absolutely should be using them for more natural language interfaces tho.

Yeah, that can also work. But I don’t see the future of software is to keep building multimillion line of code systems in a semi manual way (with or without llms). I think we will reach a phase in which we’ll have to treat code as disposable. I don’t think we are there yet, though.

We probably need higher levels of abstraction, built upon more composable building blocks and more interplay between various systems. To me that requires less disposable code though.

> but I don't think the solution is to have LLM's generate bespoke code to do it

But if the LLM needed to write bespoke code to buy the tickets or whatever, it could just do it without needing to get you involved.

seriously. I don't even wanna compile code when binaries are available in a repository. the thought of everybody preferring vibe-coding something on their own over using something that's battle-tested and available to the collective is just crazy to me.

It's more like:

- You have to work; you can't stay online all day waiting for the tickets to go on sale

- You have your agent watch for when the tickets go on sale

- Because the agent has its own wallet, it spends the 6 hours waiting for the tickets to go on sale and buys them for you

- Your agent informs you via SMS, iMessage, email, Telegram or whatever messaging platform of your choice

Yes agentic wallets are a thing now [1].

[1]: https://x.com/CoinbaseDev/status/2023893470725947769?s=20

> it spends the 6 hours waiting for the tickets to go on sale and buys them for you

Or not, because other 7 billion agents were also waiting for it.

You can already instruct AI to navigate the existing website for you and buy the tickets... OpenClaw is one such recent tool.

Personally the experience getting tickets at the moment is horrible.

Endless queues, scalpers grabbing tickets within a second. Having to wait days/weeks periodically checking to see if a ticket is available.

The only platform I’m aware of that does guarantee a ticket can be purchased if available is Dice once you join a wait list. You get given a reasonable time to purchase it in too.

So I can see why people would prefer to defer this to an agent and not care about the implementation, I personally would. In the past I’ve been able to script notifications for it for myself and can see more people benefiting from it.

It won't help, though. The scalpers have much more motivation write a better bot ("agent") than I, an occasional concertgoer, do.

Aren’t we kinda realising that disposable/throwaway stuff is, like, bad? Why do we have to go down this wasteful and hyper-consumptive route AGAIN. Can we try and see the patterns here and move forwards?

Agree in general. I don’t see how making an agent create software is more wasteful than making dozens of engineers create the same thing. The latter seems more wasteful.

We have compilers creating binaries every single day. We don’t say thats wasteful.

Well ticketmaster (for example) is used by millions of people. It seems to me like spinning up millions of LLMs to produce a million different apps is way more wasteful than having a dozen developers produce one efficient app that everyone can use?

All of the major LLMs have re-useable prompts now, so once someone makes a skill [1] that does it, anyone can use it.

Even now, with OpenClaw and all of the spinoffs, it's possible to have n agent do this today.

[1]: https://claude.com/blog/equipping-agents-for-the-real-world-...

What to use? A website where you can quickly buy the stuff you want? Or an LLM where you specify how to buy the the thing you want, wait a while, then actually do the buying, and praying in the meantime, it's not throwing your money away?

I don't know if this is the future or not, but it seems to serve no real purpose other than to enrich LLM company profits. There is real value in well designed code that has been battle tested and hardened over years of bugfixes and iteration. It's reliable, it's reusable, it's efficient and it's secure. The opposite of hastily written and poorly understood vibe code that may or may not even do what you want it to do, even while you think it's doing what you want it to do.

there is software and software. lots of enterprise software gets re-written every 2-5 years, some projects are in rubbish bin as soon as finished (if finished)

eventually people will figure out what is safe to let AI build-and-run without supervision, and what level of problem do you need to actually understand what's under the hood, audit what it does, how to maintain it, etc

I need a way to inventory my vintage video games and my wife's large board game collection. I have some strong opinions, and it's very low risk so I'll probably let Claude build the whole thing, and I'll just run it

Would I do that with something that was keeping track of my finances, ensuring I paid things on time, or ensuring the safety of my house, or driving my car for me? Probably not. For those categories of software since I'm not an expert in those fields, but also it's important that they work and I trust them, I'll prefer software written and maintained by vendors with expertise and a track record in those fields

Reply on news.ycombinator.com