Login
You're viewing the assemblag.es public feed.
  • Luca Hammerluca@social.luca.run
    Oct 17, 2026, 8:08 AM

    The term "single use software" currently lives rent free in my head.

    People use LLMs to create software specific for their needs. But neither can other people use it (hardcoded variables, no rights management, too specific), nor do they use it again (too specific, easier to produce a new one).

    Single use software is not maintained. Nobody understands it or has an interest in keeping it working.

    The creation of single use software feels cheap, but isn't. It takes time to create the software, it needs infrastructure and power. The time is experienced as short compared to how long it takes to create maintainable and reusable software. But once you take into account how many people benefit from reusable software, the single use software takes way more time to create than what it saves. The infrastructure and power costs are hidden because the data center isn't visible. The costs are subsidized by venture capital.

    Like other single use items, single use software brings risks of unforeseen consequences. From data loss or leaks to issues for maintainers of reusable software (eg. verbal abuse, less motivation and lost income).

    As an example of single use software is a German LinkedIn post of a person who slopcoded a HEMS (home energy management system). It took them more than a week. While open source projects like evcc.io/ already provide the same functionality. They could have used it for free or paid 100€ and got support for some additional devices. But they decided that they know better how it should work and now have a single use software. If it breaks there is nobody they can ask. There won't be updates and no community to discuss and develop new features. No organization will ever be interested in officially supporting it. And even if it works, it only benefits that one person.

    (This doesn't mean people shouldn't build things that only they use. But they should enjoy doing so.)

    Mit 21 habe ich Webse en weiteren Unternehmen habe ich Back up now beitet und mich auf Konzeption, Skalierung und technische Machbarkeit von Geschäftsmodellen fokussiert. In den letzten Wochen habe ich mir bewusst viel Zeit genommen, mich tiefergehend mit ChatGPT Codex und Claude Code auseinanderzusetzen. Mit Codex habe ich ein Plugin für mein Smart-Home-System gebaut, um mein Energiemanagement (HEMS) an unsere Anforderungen anzupassen. Die Software basiert ausschließlich auf Open Source: Python, PostgreSQL und Flutter für das Frontend. Das Plugin liest via Modbus TCP den Wechselrichter aus, verbindet sich mit der Wallbox, der Batterie, mit Solcast für die PV-Prognose und meinen direkt ansteuerbaren Verbrauchern. Vierzehn Stunden Entwicklung, fünf Stunden Testing, dann eine Woche Optimierung und Bugfixing - und nun automatisiert mein individuelles HEMS alles eigenständig. Früher wurden dafür hohe fünfstellige Beträge ausgegeben. Jetzt ist es einen Prompt entfernt. Individualentwicklung boomt und die Möglichkeiten wirken grenzenlos. Was mich daran am meisten beschäftigt, ist nicht die Technik. Es ist die Tatsache, dass Softwareentwicklung viel dynamischer wird und einem breiteren Publikum zugänglich wird. Natürlich bedeutet das (noch) nicht, dass sichere und performante Software einen Prompt entfernt ist. Aber man benötigt keine hohe Vorstellungskraft, um zu sehen, dass KI sehr bald bei Security- und Architekturfragen vollautomatisch leistungsfähige Systeme konzeptioniert und baut. Die Frage ist somit nicht mehr, ob man etwas bauen kann. Die Frage ist, ob es jemand benötigt und man einen Bedarf decken kann. Eines ist aber klar: Menschen mit viel Kreativität haben nun ein Tool an der Hand, das ihnen dabei hilft, technische Hürden schneller zu überwinden.
    💬 42🔄 115⭐ 49

Replies

  • 💬 2🔄 17⭐ 14
  • Not A NumberNaN@mastodon.green
    Oct 17, 2026, 9:02 AM

    @tante @luca I live in fear of the coming maintenocalypse where AI vendors take in half their income from needlessly expensive maintenance of vibed up in house tools

    💬 0🔄 2⭐ 0
  • Old Fucking Punklwriemen@librem.one
    Oct 17, 2026, 1:27 PM

    @tante @luca "Fashion" is an apt description for software development, and "fast fashion" fits the slop coding. I've often referred to software development as a popularity contest due to the lack of measurement. Developers will follow trends without asking why or to what benefit; just look at all the misuse of Python these days. From the reports of slop projects, I'm sure that actual measurement would put slop use in the avoid column of the process spreadsheet.

    💬 0🔄 1⭐ 0
  • L'égrégore André ꕭꕬMabande@mastodon.social
    Oct 17, 2026, 8:31 AM

    @luca I honestly understand the urge for a new setup if he's paying tens of thousands for the old one (that's what the "fünfstellige" bit is, right? My German's poor to say the least), but I really don't understand why he chose a week+ of slopping instead of a week of research to see if there already was a cheap/free and maintainable solution available.

    💬 2🔄 1⭐ 0
  • 3mind3mind@freiburg.social
    Oct 17, 2026, 8:45 AM

    @Mabande @luca He did not say, he had paid such an amount („fünfstellige Beträge“). He only mentioned, „in the past such an amount would have been paid to develop such a system“. But you‘re right: Either he should have researched more thouroughly to find available open-source tools doing the same (but much more tested + community support) or pay some hundreds to buy a commercial appliance HEMS. But now he only has a „cheap“ proof of concept that „such thing is possible with LLM“ with no support.

    💬 1🔄 1⭐ 1
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • NoisecolorNoisecolor@toot.community
    Oct 17, 2026, 8:52 AM

    @luca
    That's silly. You would expect to check first and at least fork working project or something.

    Single use software is an interesting concept. Because if you extrapolate to a future with much better and cheaper llms, maybe that is the future. Only developers can really have a community and support and collaborative work. Normal people will just say what they need.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 2⭐ 0
  • DrYakdryak@mstdn.science
    Oct 17, 2026, 9:11 AM

    @luca and in the act of having a slop machine extrude a piece of single use software, that person gained no experience, no domain knowledge, etc.

    Even if that specific problem is niche, you never know where some of the acquired know-how could become handy.

    Maybe researching opensource tools and duck-taping a couple of them would turn out useful.

    Instead, that person just wasted time arguing with a word extruder software.

    [...]

    💬 1🔄 3⭐ 1
  • DrYakdryak@mstdn.science
    Oct 17, 2026, 9:23 AM

    @luca […]

    counter example:
    in 2018 i wrote some scripts to automatically backup my Linux smartphone onto my Pi server.
    A mess of rsync, ssh forced-commands, and bash scripting ducktape.

    in 2020 when the pandemic struck, this pile of scripts formed the base of the automation harness that runs automatically the analysis of genomic sequencing of viruses for the surveillance in Switzerland.
    We processed around 100'000 clinical samples, and are still doing environ. surveillance in wastewater

    […]

    💬 1🔄 2⭐ 4
  • DrYakdryak@mstdn.science
    Oct 17, 2026, 9:26 AM

    @luca […]

    the first piece of software is pretty much this type of "single use software". It's very specific to my usecase (Linux smartphone, Pi).

    But the problem-solving skills involved in hacking that solution turned out super useful in an unrelated but seriously important settings.

    And the only way to hone one's problem solving skills... is by acutally solving problems, actually doing stuff.

    Arguing with a word extruder won't help.

    […]

    💬 1🔄 3⭐ 1
  • DrYakdryak@mstdn.science
    Oct 17, 2026, 9:29 AM

    @luca […]

    Those slop-machine are actively robbing us from the opportunity of creating the next generation of good problem solvers.

    (And in the same go are also depriving the "AI" big corporation of the source of expertise they scrape and rely on to train their next version of slop machines)

    […]

    💬 1🔄 2⭐ 1
  • DrYakdryak@mstdn.science
    Oct 17, 2026, 9:30 AM

    @luca […]

    (NOTE: I am not saying that this person should "learn to code" and write their own software from scratch in stead of vibe-coding.
    Problem-solving could also be researching and using whatever opensource tools exist, and learn to configure them to the specific needs)

    💬 0🔄 1⭐ 1
  • Holgintheairholgintheair@det.social
    Oct 17, 2026, 9:12 AM

    @luca
    This guy spent more than a week slopcoding something I managed to install in 2 hours using an existing foss project (evcc).

    💬 2🔄 1⭐ 0
  • 💬 0🔄 1⭐ 2
  • Arjen P. de Vries Timmers 🕊️arjen@idf.social
    Oct 17, 2026, 2:56 PM

    @holgintheair @luca which raises the question why the coding assistent did not recommend building on evcc - they are still not that good in capturing the end goals.

    I've always enjoyed writing Perl oneliners (in the late 90s), programs one should think of as "write once code". (Perl is cryptic but can be exceptionally useful.)

    The real problem starts when you, one day, want to reuse such a "throw away" program. And: it's really hard to predict you will never again need some software solution.

    💬 0🔄 0⭐ 0
  • Gabriel Pettiertshirtman@mas.to
    Oct 17, 2026, 9:13 AM

    @luca I was thinking about the parallel with visual basic apps and excel / access databases of the 90s/00s, that was a popular way to build throw away / bespoke software, and over time, a lot of that was replaced by more professionally developed apps, sometime after great pain, but also, a lot was just forgotten as soon as they outlived their usefulness.

    I think the current wave of bespoke LLM software, is interesting, and the usefulness and quality is certainly superior to what people could cobble together with limited understanding with these tools, but if we mistake the apparent polish, for quality, and give too much responsibility for this software, outside of more professional audit/support/management, there will indeed be great pain.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 1⭐ 0
  • NewherehowNewhereish@mastodon.social
    Oct 17, 2026, 9:42 AM

    @luca There is a problem with this analysis. There is nothing wrong with single use applications, and time is valuable too. If it is faster for the end user to vibe code up a single use thing for their specific need, that is time spent getting the thing done, using it what it is to be used for, and moving on. It saves the most precious resource humans have, time.

    💬 1🔄 0⭐ 0
  • Nicole ParsonsNpars01@mstdn.social
    Oct 17, 2026, 11:28 AM

    @Newhereish @luca

    That "time savings" for a few is coming at a very high price to everyone else.

    tomshardware.com/tech-industry

    Like many modern conveniences, we often ignore the full supply chain costs and the environmental impact.

    Data centers pollute & waste energy; just so a few billionaires can get rich off "convenience".

    wired.com/story/chevron-willia

    Fossil fuels are toxic products that must be phased out; the AI bubble is thwarting that switch
    thehill.com/policy/energy-envi

    cnbc.com/amp/2026/06/22/chevro

    💬 1🔄 2⭐ 0
  • NewherehowNewhereish@mastodon.social
    Oct 17, 2026, 7:41 PM

    @Npars01 @luca Time is the only resource that can only be spent, it cannot be conserved. Any reduction in its use is therefore of unlimited value. There is a reason convenience has been, is and will forever be king.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 1⭐ 0
  • 💬 0🔄 1⭐ 0
  • d1d1@autonomous.zone
    Oct 17, 2026, 10:31 AM

    @luca yeh it's so real holy shit. i'm seeing it happen all over the place now

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Tim Ward ⭐🇪🇺🔶 #FBPETimWardCam@c.im
    Oct 17, 2026, 10:51 AM

    @luca Once Upon A Time we had a concept of "temporary" software, which didn't have to be built according to our normal engineering standards / quality rules.

    Typically this would be a tool for a one-off data conversion, something like that. It had a lifetime of 24 or 48 hours, forget which, after which anybody could delete it / throw the paper tape away / whatever.

    💬 0🔄 0⭐ 0
  • Sir thalon :klingon:thalon@embassy.social
    Oct 17, 2026, 11:26 AM

    @luca I understand your line of reasoning, and you do make some valid points. Nevertheless, I would like to partially disagree. In my own experience, my software needs often fall into the "long tail"—meaning that the software I require would never be economically viable because the concept doesn't scale. Under a conventional paradigm, such software would likely never be built in the first place. With AI, however, I can tailor solutions to my specific needs, which represents a huge step forward for me.

    💬 0🔄 1⭐ 0
  • Philippa Cowderoyflippac@types.pl
    Oct 17, 2026, 11:35 AM

    @luca Single-use software is one of the reasons I will only rip so heavily on Python (admittedly in terms of my own comfort, pattern matching added a lot)

    The days of bashing things together with awk and sed are gone even if you know what's good about those tools, and there's a lot of convenience features for <KLoC programs that don't make sense to dedicate the syntax to in other languages!

    💬 0🔄 0⭐ 0
  • Philippa Cowderoyflippac@types.pl
    Oct 17, 2026, 11:36 AM

    @luca (The thing is, some single-use software exists because the big alternatives take long enough to configure that with the right background you might as well have written it yourself

    Then you get the unholy confluence that is "we don't need approval for Excel"...)

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Julik Tarkhanovjulik@ruby.social
    Oct 17, 2026, 12:02 PM

    @luca I think we are looking at it from the wrong angle. "A system already exists for this" is a projection of "Why didn't the guy just use <$my_thing>" which makes us all feel bad, but we haven't really checked "why is contributing to $project seen as a drag by commoners". OSS has created a very elitist attitude towards either newbie users or driveby-contributors, and it is chock-full of design decisions which are - let's call it for what it is - bad. One natural way out is vibe.

    💬 0🔄 0⭐ 0
  • Michal BryxíMichalBryxi@mastodon.world
    Oct 17, 2026, 12:15 PM

    @luca I mean yes, but also no. The term “maintainable” is a never-fulfilled promise of IT. Ever. One gets incompatible upgrades. One gets vendors breaking their APIs. One has to monkey patch an integration because who would expect anyone to follow any standards? Etc.
    So while I do agree that the Single Use Software is *mostly* for one user, it also allows something we’ve never had before: Maintenance-less systems. Assuming the human is out of the loop. Completely.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • christian mockcm@chaos.social
    Oct 17, 2026, 12:55 PM

    @luca I was gonna start arguing with you, but your last paragraph took the wind out of my sails.

    Because I did the same thing, built my own HEMS integration, with about the same effort, but I did it with my own brain, and I learned *a lot* (and I am not a professional programmer, although I've been coding on and off for 4 decades).

    But going from "works well for my environment" to something that could be reused by others is a lot of effort, needs cleaning up, abstracting things, documenting /

    💬 1🔄 1⭐ 0
  • christian mockcm@chaos.social
    Oct 17, 2026, 12:58 PM

    @luca and that's where I was about to start arguing, is that worth it in any case? Be it one-off shell scripts for a specific task that I might dig out in two years time and use as the base for a similiar task, or bigger systems such as that HEMS. Because there's also danger in over-generalising stuff, going down the rabbit hole and never solving the original problem nor producing something that's easy to reuse for others.

    Does that mean I approve of people vibing together single-use solutions?

    💬 1🔄 0⭐ 0
  • christian mockcm@chaos.social
    Oct 17, 2026, 1:00 PM

    @luca hell no. Because one doesn't learn from them, doesn't understand them, can't really see what's going on or reason about correctness, or fix bugs (in the sense that one understands why they occur).

    💬 0🔄 0⭐ 0
  • Mike FulkFulk_It@mastodon.social
    Oct 17, 2026, 12:59 PM

    @luca Some people have to learn things the hard way. They're unable to look and reason around them, and see how things will turn out.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Old Fucking Punklwriemen@librem.one
    Oct 17, 2026, 1:17 PM

    @luca "Single use software" is a good observation, also known as "throw-away code" (which never gets thrown away). Capers Jones has expressed excitement about slop, because of the "potential for reuse". I wonder if he's started measuring any slop coding efforts. I think he's going to change his mind. He should know better than to jump the measurement step.

    💬 0🔄 1⭐ 0
  • Stefan Ihringercompfu@mograph.social
    Oct 17, 2026, 1:18 PM

    @luca Vibing your own solution is also a way to never have to think about whether your workflow makes sense. Say, you want a customer relationship database. If you look at existing software, none might fit the way you have been working so far with just Excel sheets or whatever. But maybe your workflow was inefficient and you've just entrenched it.
    Existing software might work differently because people with more experience have concluded that it makes more sense to do things in a certain way...

    💬 0🔄 1⭐ 0
  • 💬 0🔄 1⭐ 0
  • EvoScaleEvoScale@c.im
    Oct 17, 2026, 1:30 PM

    @luca Copyright issues are now accelerating, considering the next level hacking available to #AI and #TechBros #AbuseOfPower.

    Just because your average coder and end user sets aside a smaller, simpler code set, doesn't mean it's not easily archived and quickly (read: at the speed of the fastest AI available) aggregated into newer agency, outside of any human intervention.

    So does that leave any equitable accountability, solely up to the most powerful AI, while the .01% wealthiest remain greedily, and blissfully, ignorant?

    💬 0🔄 0⭐ 0
  • blausand 🐟blausand@chaos.social
    Oct 17, 2026, 1:32 PM

    @luca This is quite a weak take, seen from my perspective.
    I regularly develop one-shot tools and often it took me ages, but I enjoyed (point given) the learning process.
    What I experience: The vibe-coding era brings a bigger pile of shame when it comes to maintenance and doc. But that's a well-known cultural challenge and AI helps a lot with that overall, too.

    I generally refuse the conclusion AI erodes our mindful processing; it happens a lot now, but I'm confident our habits will evolve.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 1⭐ 0
  • dusoftdusoft@fosstodon.org
    Oct 17, 2026, 1:50 PM

    @luca Single use software has existed before, think scripts and patches and hacks. But indeed now time to prototype is 30 mins, so lots of software ideas can be realized quickly (and then either developed further properly or forgotten). Nothing wrong with that, on the opposite. But indeed vibe coders without programming experience forget the maintenance debt. Programmers on the other hand can be more efficient in architecting systems and not getting buried in the details.

    💬 0🔄 0⭐ 0
  • Seth Mosdatabeestje@noc.social
    Oct 17, 2026, 2:10 PM

    @luca I think of different software, with a single registration code unique per PC and you get to buy it again.

    I despise that software

    💬 0🔄 0⭐ 0
  • Q.E.DanQEDan@mastodon.online
    Oct 17, 2026, 2:54 PM

    @luca Single use software isn't new to LLMs. It's been around since the dawn of computing and remains extremely prevalent in forms like spreadsheets and shell scripts, whether prepared by hand or with LLMs.

    💬 0🔄 0⭐ 0
  • mbkmaybe_mbk@mastodon.social
    Oct 17, 2026, 3:06 PM

    @luca

    I would separate three things: actual single use software (not a new concept, a lot of scripts are basically that), software only for one person's need/use, and then whatever vibe/cuck coders do. A lot of generated stuff is "single use" only because they have to replace it fully when it fails. 1 and 2, on the other hand, are a different story.

    (And pray to your gods or some other person's gods when LLMs start concepts for architecture and security, given the data they were trained on.)

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • ambiguous0ambiguous0@mstdn.plus
    Oct 17, 2026, 3:32 PM

    @luca there are so many reasons to get upset about LLM coding, but people creating niche apps for themselves is not it. This is pure elitist nonsense.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • AyCodeaycodeai@mastodon.social
    Oct 17, 2026, 8:04 PM

    @luca A habit that would have saved that week: before an agent builds anything, ask it to list existing open source projects that already do this and why they don't fit. Often the most useful prompt of the session.

    💬 0🔄 0⭐ 0
  • ·J Moppjmopp@masto.ai
    Oct 18, 2026, 7:03 AM

    @luca "single-use software" back in my day it was a script I threw together in half an hour

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0