Thoughts on AI usage for retro projects

[Rhetorica]

Below are my honest opinions on AI and the use of AI in this community. These are not divine edicts and nobody has to agree with them. Please feel free to engage with them or post your own thoughts in this thread.

I would like this thread to be the one-stop shop for newcomers who want to get a feel for how the community regards AI, so if you do not see your stance represented here, please consider sharing it.

1. Artisanry doesn't need to be necessary to be worthwhile
Whatever we do with AI code, we should acknowledge that hand-built software deserves to be exalted and recognized for the human effort that went into it—and into learning how to make it. Previous is a marvel of development, and @andreas_g's talents in stewarding that codebase—and the work of his contributors and predecessors, and the upstream work of the UAE and Hatari teams (among others)—can never be replaced by some cynical kid asking a genie to make the same thing. It is admirable and commendable that Andreas has made the decision to keep the codebase AI-free, though I think we all recognize that there is little room for innovation when it is already nearly feature-complete and (to my knowledge) bug-free.

But at the same time, this celebration of genuine human labor does not mean we cannot also have conveniences. Handwoven cloth today is still valuable and commands a premium, even if we have automatic looms. The armies of women and men who were forced to perform this labor to make ends meet did, eventually, move on to better and more fulfilling things. Most cloth we use on a day-to-day basis has become more ephemeral—often a disposable commodity—but this lower-value material has not led to handwoven textiles being devalued in the same way, even if it is delicate or imperfect.

2. AI is already in use, and this is irreversible; if we ignore it, we will cede the world to barbarians
LLMs are capable development partners at this point, whether we use them for little bits of advice or to plan out an entire system. They do, of course, struggle with human sensibilities still—they do not know what makes a game fun and they cannot stumble onto new UI paradigms—but these are, if anything, comforting limitations, as it means our input still has value in the process.

Moreover, people have noticed that this is the case. We are past the Ned Ludd moment: AI usage is not going to stop globally. The environmental impact is an inevitability, though perhaps if the trauma of high oil prices lasts long enough, we will finally be weaned onto renewable power for purely geopolitical reasons. The job market impact is equally inevitable, though perhaps not as quickly as the hype-men and tech-bros promised. We cannot destroy the Torment Nexus, but we can still steer it. (The specifics of that conversation, however, are perhaps for another place and another time.)

With these things in mind, the only real power we have as a community is to choose to either make the best of AI or to boycott it on principle. As a Canadian, the zeitgeist this past year from my perspective has been that walking away from a messy situation is merely leaving it to fester. If we don't use AI tools, we lose productivity, and only the people without scruples are productive. Therefore if we disengage, we cede the public square to grifters and hacks who have no trouble amplifying their message, and they'll always be able to do a shitty job a million times faster than we can do something beautiful by hand. Either way the majority of new code being produced will be AI-generated, but if we participate, at least it won't all be the worst possible slop. We cannot stop AI usage from becoming normalized, but we can stop lazy AI usage from becoming normalized.

3. Open source projects have successfully negotiated AI usage
Though I've mentioned code numerous times so far, the above really only broaches general concerns about LLM usage that are not specific to software development. As some of you are no doubt aware, there are two key threads of conversation that have happened around AI usage within the open source community around AI usage.

The first of these is an anxiety over an AI regurgitating large patches of source code of an incompatible license due to a relatively small training set. There are many reasons why we should not take this concern seriously:

• When a contribution is submitted to an open source project, its provenance is rarely a subject of concern. We have not, historically, been concerned about the risk of plagiarism. The only times open source projects have actually worried about this sort of thing is when they are meant to be clean-room reimplementations of something that has leaked, such as when ReactOS was forced to conduct an audit in the wake of the Windows source code leak. It is not, at this time, a realistic legal threat to small open-source projects with no revenue streams, especially if the offending fragments are so small they can simply be rewritten and replaced.
• The a priori assumption that LLMs will habitually produce text fragments of a copyrightable size from a single source is predicated on a lack of knowledge about how they are actually trained. The programming capabilities of modern commercial LLMs like Claude 5 Fable and GPT-6.0 Astra are fostered through in large part through writing tests and using automated tools to validate the quality of the code. Because of this, these models actually have to learn structure at a deeper level, and improvise. Even if the training data includes a valid solution to your prompt, it is unlikely to meet Claude's stylistic and structural preferences—unless, of course, Claude already generated it, in which case the code is not copyrightable in the United States, and none of these concerns apply.
• The more general moral objection—that Anthropic, OpenAI, Deepseek, Moonshot, et al. harvested data without respect for copyright—presupposes that copyright law is reasonable in the first place. The truth of the matter is far messier: copyright overwhelmingly serves big corporations, history is replete with examples of technological revolutions that were kicked off by jubilees of copyright, and we wouldn't need open-source licenses at all. The GPL was created to stop the cannibalization of the commons; AI is not doing that. Indeed, humans also learn by reading and gaining experience from copyrighted lesson material, textbooks, and source code, but our use is—rightly—considered transformative.

The second concern is about whether or not contributors can take responsibility for their code. Many projects have been inundated with low-quality contributions by bots that have been told to "go improve some open source projects" or have otherwise been set to the task. GNUstep recently had a problem with a new contributor who suddenly started submitting dozens of patches, not all of which worked. But this is quite simply an etiquette problem and has little to do with the technology involved; any team would object just as readily to an outsider mass-submitting changes that did nothing but beautify comments or dump a bunch of typos into the codebase. We should not be surprised that a newbie who previously did not have the tools to contribute whatsoever has a lack of awareness about the culture around them. The question of what to do about children with chainsaws is a purely social one and is unlikely to last forever, as management strategies and general awareness of the problem spread.

So, yes, contributors should be able to take responsibility for their code; this has always been a requirement. But if your project team has a size of 1, this does not matter.

Both Debian and the Linux Kernel eventually came around to these conclusions, and the GNUstep Project is not far out from accordance with them as far as I know. If Linus Torvalds, of all people, is willing to tolerate responsible AI usage, I would be hard-pressed to find a more compelling argument from authority. (The target on his back—i.e., the potential payday—will always be much larger than anyone else's.)

(This is where I'd insert some boilerplate acknowledging the suffering of Alex Perez, whose work on the ZuluSCSI firmware was appropriated by the BlueSCSI team without due credit or compensation. Some scumbags don't need a God in a Box to express their scumbag-nature.)

4. For most obsolete tech, there are no mentors
When a software ecosystem is current, people don't just use books to learn; they have access to electronic documentation and an active community to ask questions. Those resources are scant or non-existent for NeXT and OpenStep APIs, even though at the time NeXT's Digital Librarian references were state-of-the-art. When NASA set out to rebuild the F-1 engines used by the Saturn V, the loss of knowledge in the intervening five decades was so severe that they had to reverse-engineer museum pieces starting from first principles. (This may have contributed to the SLS's schedule slips and budget overruns that led to the program being canceled in favor of the Atlas V, which uses a Russian-made engine for the lower stage.) Having the AI around fixes those gaps, especially the gap of human expertise; even when veterans exist, they probably haven't used most of these APIs in decades, or have re-learned how to do things 5-10 times as the tech evolved under them.

On a personal note, getting quick access to the synthesis of obscure reference material is probably the single most useful thing that AI has done for me. To produce the rationalized SVGs for the recent thread about the NeXT logo, Claude performed hundreds of statistical analyses to find the best-fit model for the number of modules that Paul Rand's drawings used. This was work I could have done myself (I have a graduate degree in statistics modeling for computational biology, and extensive graphic design experience), but wrestling with a stats library in R or Python would have made the project take days I didn't have to spare, and the work was not creative: the goal was to recover the logo's original metrics, for which there was a correct answer. I expect that, once I finally can sit down and start writing Objective-C for myself in a few months, this kind of time-saving will continue to help me learn.

5. The silence has been deafening
In the two decades before this past year, very few new projects were undertaken to port software to NEXTSTEP or OPENSTEP, much less to write new programs. The majority of these were drivers for virtual machines, or ports of historic, non-ObjC software from the early 2000s before the adoption of C99 and C++03 features made backporting too painful. At the very start of this period there was maintenance of historic NeXT software like XS3mod and TIFFany, but for nearly two decades there was near-total silence; to my knowledge the only new application in that timeframe was @jeffburg's MathEdit, which was a heroic undertaking to span the whole breadth of OpenStep/Cocoa's history, but is, also, kind of just a neat toy, and not exactly a must-have for daily driving a vintage computer.

In short: if it were straightforward to develop new NeXT software, there would have been more than this thinnest trickle of output. But it isn't easy—in fact, developing modern software for modern machines is a lot easier, thanks in no small part to the innovations NeXT pioneered or helped to pioneer, and also the mundane realities of digging deep into old material where expertise is lost. Only in the most recent period have we seen the floodgates open, with multiple passionate contributors spending months working on building new drivers for real hardware, new build systems, and new solutions to old usability problems.

It is true that there are retro communities where the music never died—for example, the C64 demoscene has remained vibrant and full, and it would only be natural for a scener of 30+ years to look down upon someone asking an AI for mentorship instead of a human. Every community, therefore, needs to make its own decisions about whether automated methods are normal or abnormal, and we might expect that—someday far in the future, when AI usage has reached an equilibrium—fledgling resurrection communities will start out AI-heavy and gradually wean themselves off it, purely out of pride in their craft, as experience grows.

For the foreseeable future, however, I do not think we have anything to gain from stigmatizing AI usage in this community; indeed, we have a great deal to gain from encouraging it. So, I say we should let the thaw continue.

[ZombiePhysicist]

3 comments.

First, bully. Here here. 3 Cheers and all that. Well done.

Second:

"(This is where I'd insert some boilerplate acknowledging the suffering of Alex Perez, whose work on the ZuluSCSI firmware was appropriated by the BlueSCSI team without due credit or compensation. Some scumbags don't need a God in a Box to express their scumbag-nature.)"

While I agree with the above, I'll note any prior tsk-tsk snark directed at someone's "save the @" are now declared null and void!

Third:
AI is a tool. It can be used to create abhorrent disastrous problems. And it can be used by good. Most time its the character and intentions of the people that wield it that matter most. It's not a right or a wrong, it's a design choice, and as most design choices, they come with pros and cons, that are best left to the characters wielding the tools.

TLDR, agreed.

[stepleton]

The participation-magnifying effects of AI on the NeXT community are also found in the Apple Lisa community, where there has been a similar process of reckoning. My contribution is here. But I'll quote the excerpt that seems most general (and it also shows why my NeXT interest is a bit less fervent than my other retrocomputing pursuits, despite my lingering here):

To me, I like the old computers I collect (my Lisas, the PERQ, the Tektronix vector-display machines, the transputer setup, etc.) because they feel like windows into a different universe where a different set of ideas took off to the ones we know today. When I use or write code for these computers, I'm often trying to imagine being in that universe and seeing what it's like. I collected a few NeXT machines in the late '90s with that same "what if" speculative enthusiasm. Then OS X took off and NeXTs became a lot less interesting to me. We found out the answer to "what if" --- it's this universe where we are right now!

For now, AI coding doesn't trade so much in "what if". If you have a program idea in mind for the Lisa, it will just go straight ahead and make it, overcoming any obstacle or point of incongruence between your idea and the underlying libraries and tooling. But those obstacles are traces of the realities, preferences, and philosophies that made the Lisa what it was when it was introduced. The people who created the Office System imagined a computer that worked in a certain way, and the ecosystem that they started would have evolved with those visions as a major influence. (Of course those visions were "wrong" --- that's at the root of why the Lisa failed. But what if?)

If I make a Lisa program, I want to run into those frictions that come from 40 years of divergent history and not being on the same wavelength. I want to try to get into its designers' heads. I want to have those moments of "so that's why it's this way". I think this kind of understanding is really helpful to trying to make something that would be a convincing "what if" Lisa program from, say, 1985. Could I do this kind of exploration by using AI to code? I actually think I could! But I'd have to be pretty careful to rein in the AI and keep it patient and "true" to the feel of the experience, and to allow me to discover along with it. AI coding agents are not made for patience --- they are built to go relentlessly and just get it done, no matter what "it" is. So you need to do work if you want to ride along with them on the scenic route.

Of course, that's my own preference and my own kind of enjoyment: it's not for everyone! And it won't take long to count the number of Lisa programs I've made since the dawn of capable coding agents: just one, my LisaVox demo, which I did by hand for the fun of it. And that was January. Coding agents have got a whole lot better in the eight months since then!

I'm totally in favour of people chasing their own pleasure and visions for Lisa programs and capabilities. AI systems have clearly revolutionised this: we went from maybe one or two Office System programs per decade to, like, ten in the past few months. People are dusting off plans and hatching schemes for more, and AI is making it possible. That's great! The only thing I'll say is that these new plans are inevitably going to evolve the ecosystem. @bmwcyclist's new utilities are great examples of this: they're setting up ways that people manage files on Lisas, and at some point they may become a big part of the "standard way to do it". Are these approaches the standards we want to wind up with? What are their pros and cons? I'm not hinting either way; I really have no opinion at the moment. What I would say though out of naked personal interest is that I hope we don't evolve in a way that makes it more difficult to achieve my "what if" fantasies someday. I have a few ideas for how that might happen (e.g. new convenience and interoperability libraries that start to subsume the original ones offered by the OS/ToolKit/Office System), but I've gone on long enough, I think! If anyone has read this far, I hope it's been interesting...

[onionmixer]

Thank you for bringing up such a great topic. As a newbie to this community, it's not an easy subject for me to just dive into casually.

Like many of you in this community, I'm someone who has "one foot" in programming. I've been programming for quite a long time, spent nearly 20 years as a PM, and because I worked with Linux for so long, my career as an SE actually ended up being longer. Naturally, I drifted quite far away from the hands-on work of building software. Though there was a time when I even built my own Linux distribution

When I first started using AI back in September 2025... I genuinely laughed.

"Why would anyone use this? It's barely at a level where it can do anything outside of basic web coding."

What changed that mindset—from my perspective—was around April 2026. Claude's progress was so immense that I could clearly divide my work into "before" and "after" that point. By around June 2026, Codex had also improved enough to be reasonably useful.

To be honest, my economic situation and the resulting family pressures became so severe that I went through psychiatric counseling in 2025. Things had been continuously tough ever since the pandemic... but 2025 was—truly, insanely difficult (and things are still tough now). But... working with AI genuinely... really improved my condition a lot. Moving past that period where I felt like a useless person in this world... now, I've reached a point where any project I want to build can be brought to life with AI. Of course, it requires paying some AI subscription costs and grinding away my personal time, but if anything, it seems to be working out for the better in my case.

Naturally, when I first used AI, the output I produced was far more pathetic compared to colleagues who had lived as programmers for a long time. It was just at a level for my own personal use. But now, about a year later... I'm turning out results good enough that those very colleagues ask me to teach them how to use AI. Managing servers, web programming which mainly generates revenue, and the "unprofitable, hobbyist projects" I do on the side... I'm producing such insane results that I run out of tokens every week and have to ration them. My recent classification was PM (Project Manager), and I'm simply doing the same thing with AI that I used to do by dividing tasks among people and checking up on them.

Now, let's talk a bit about the realistic side of using AI for the work "I" want to do.

Most of the projects I want to tackle are generally not things that make "money." When I built a mobile app connected to a web service 12 years ago (which completely tanked because of the pandemic—only the web backend survived, haha), I redesigned the encryption protocol between the mobile app and the server daemon based on a few SSL algorithms. And I built the server in Objective-C. Does building something like this make money? No, it doesn't. I knew full well that just using plain old HTTPS would have been much easier, but I persuaded my team members and built it this way anyway. Thanks to that, I got a mobile app with a fairly guaranteed level of technical completeness and stability, but! That was pure self-satisfaction. It was just a matter of "pulling off" what I wanted to do, not an outcome that generated "money." The only reward was hearing people around me say, "Whoa... you guys are out of your minds."

Recently, as I prepared to reuse some of the libraries from that project and open-source them, I converted the core of the server daemon from Objective-C to C++ (because I figured that having to set up the server environment every single time just to compile libObjc2 would be inconvenient for others. Most regular engineers rarely enjoy dealing with such tedious setup processes). This task? Naturally, the engineer didn't want to do it. So I handed the conversion over to AI... After about 3 weeks, a prototype came out. Once I laid down the major architectural lines, that engineer—who hadn't really believed I could pull it off—is now using AI like crazy and applying it enthusiastically to his own personal projects

Back when I made a living through programming, I used to be deeply obsessed with development methodologies. eXtreme Programming, and Test Driven Development—known as XP and TDD. What is one of XP's signature practices? Pair Programming. Working on all sorts of tasks with AI, I've come to think so often that this might just be the ultimate partner and setup for applying these exact methodologies. A human can watch the AI's output in real-time, pointing out when it's heading in the wrong direction and correcting it on the fly. And through having the AI write countless unit tests—which honestly I wouldn't have bothered doing before out of sheer laziness—I can significantly elevate the quality of the final output. Plus, as Martin Fowler talked about, I discuss refactoring with the AI and refine the technology, 15 hours a day... Could I ever do this with a human? I'd love to, but in most cases, people just can't operate like this. They might be more skilled than I am... but they simply don't enjoy the sound of a computer fan whirring as much as I do.

I know that this approach has helped keep my mind and mental state in a comparatively healthy place. The current version of me feels infinitely better than the miserable past version who relied on psychiatric meds for stress, lived in a daze all day, just getting by and breathing. Minor stresses don't phase me anymore. My general relationship with my family has improved vastly, too, because I can be relatively more patient with my kids.

Let's also talk about hobbyist coding. Coding that doesn't make money... previously required hiring an engineer or begging someone else to do it. What I thought was fun might not have been fun for others. And if they turned me down? Even if I had the money, it simply couldn't get done. Nobody was going to volunteer to build a VGA driver for OPENSTEP, and 12 years ago... even if I had tens of millions of won, I couldn't find anyone to hire. But AI doesn't "mock" the work I want to do; it sits down and brainstorms paths to make it happen together. It may not have the profound insights of a senior engineer, but when I propose a direction, it reviews it and judges whether it's feasible or not. Isn't this XP and TDD? I want to pour my absolute all into it... and as long as I put in the effort and keep pointing the AI in the right direction, it sticks by me and pushes through to the end. And when a job finishes successfully, it even praises me, saying, "Your persistence brought this result to life" (though I'm not foolish enough to miss the fact that it's just a routine pattern, haha).

And when working with humans, it's been quite a while since I last felt this kind of dynamic. Compared to regular, professional engineers who half-ass their work and lack genuine passion, collaborating with AI feels infinitely more serious and rewarding. At least to me.

Discussing how to implement OpenGL acceleration in OPENSTEP, investigating memory-blocking regions, uploading data to VRAM for acceleration—yet when the AI doesn't know how to render it, guiding it to map the computed VRAM data into Mesa and directly reading the related memory to draw it onto AppKit... This kind of high-level architectural guidance is still something a human directs (for now, at least). A typical human engineer... isn't going to work this hard for 220 dollars a month. They won't touch tasks that don't look good on their resume, and even if you offered them an annual salary exceeding 100 million won to build a driver within a month? No way... But AI—whether a task makes money or not—responds with genuine sincerity. When you point it in a new direction, it doesn't complain; it starts by doing the research. If this were a commercial project where an engineer was being paid a fair market rate, I wouldn't do it this way either, because I'm lazy too. But... I'm not well-off enough to burn that kind of budget on hobby projects. If it's something I want to do even if others laugh at it... in cases like this, AI is infinitely better than a human.

I collect all sorts of "hardware"—from the Apple II and MSX to modern-day PCs and various gaming consoles. Every single piece holds memories and sparks the curiosity I had as a kid staring at computer magazines. My dream is to breathe my desired "usability" into all of these vintage systems, giving them a "purpose" so they aren't just relics of the past, but remain meaningful even if used today. To achieve that, instead of hiring people and waiting around in vain... I need a companion who will simply charge forward in the direction I want. And whether that companion is a human or an AI... doesn't really matter to me. As long as I'm in a healthy enough state of mind to communicate with others, produce results, and—if those results turn out great, wonderful; if not, so be it—it stands as proof that I am alive. Whether I leave that proof behind through AI or through my own labor... doesn't matter "to me." As long as the outcome of something I built remains as a useful record to help someone else out... that is more than enough for me.

This is just my personal view, but... we've reached a reality where whether you used AI for "coding" or not has become irrelevant. However, "what to build" using AI still has to be decided by a "human." No matter what we build, if the end user of that creation is a "human," the judgment of "what kind of outcome will actually be useful to people" can only be made by a human. Of course, there are people who find pure joy in the coding process itself, and I fully respect their perspective. I think the feeling I used to get from programming back then was truly wonderful.

However, around 3 years ago... after an acquaintance I was very close with passed away (he was about 4 years older than me), it gave me a lot to think about. I realized that if I kept going the way I was, I wouldn't even be able to start the hardware collecting and the things I had only been "thinking" about doing before I died. Around that time, I started using AI, and as AI has grown increasingly capable recently, I've caught a glimpse of hope: "Maybe, while I'm still alive..." That's why... as long as my health and condition permit, I'm not going to stop. I can build small programs, but I don't have the chops to analyze the BSD, Xorg, or Linux kernels and write device drivers. In my own domain, I have enough skill to self-certify, but it's a modest level when it comes to realizing all my dreams—and I'm self-aware enough to realize that other people aren't going to step in and build these things for me.

This got a bit long, but... regardless of what others might say, I fully intend to keep using AI enthusiastically for the projects I want to pursue. If having too many things I want to do is a problem, then I suppose that's my problem... but other people's perspectives do not take priority over "my mortality." For similar reasons, I recently left the Reddit community "https://www.reddit.com/r/vintagecomputing/", which is how I discovered this NeXT community. Some people there simply fail to distinguish between AI-generated garbage content and the actual value generated by AI. For the record, my final post there was blocked by a moderator, so I just went ahead and posted it separately in a new Reddit community I spun up myself.


https://www.reddit.com/r/WorldLikeAnOnionmixer/comments/1wsi4gi/the_following_is_the_final_post_i_wrote_in_the/

I also agree that the series of tasks I'm doing in this community might give some people a negative impression. Therefore, if there are those who disagree with my views, I can fully understand and respect their perspective as well.


P.S. Since I'm not very fluent in English, I wrote the entire post in Korean first and translated it into English using an AI tool. If there happen to be any expressions that sound unpleasant or offensive—contrary to my actual intent—I kindly ask you to keep that in mind and take it in stride.

[ZombiePhysicist]

We certainly live in interesting times. By some counts, both at the precipice of the darkest of times, and also at a time where the most creative have the greatest lever in bridging their imagination into reality.  A bizarre duality.

[onionmixer]

We certainly live in interesting times. By some counts, both at the precipice of the darkest of times, and also at a time where the most creative have the greatest lever in bridging their imagination into reality.  A bizarre duality.

Nevertheless, my personal stance is that we just have to do our absolute best in our current situation. Honestly... aside from that, there's really nothing else someone like me can do, haha.

[Adam_Hall]

Despite the best efforts of tech companies, I suspect all this AI tech will be pretty much free and their efforts to gate keep it will eventually fail spectacularly, like AOL tried to gate keep the Internet 20+ years ago and failed. It is going to take a drop in hardware prices and an end to the deals they are making with NVIDIA to keep hardware expensive, but I hope to just run everything locally and see a future where that will be the most common solution for individuals and companies who want to use AI. I rent a pod right now and do little experiments with open source engines.

[ptek]

Yes AI is very useful, scrolling through a 200 or 500 page specs pdf is not fun

Waiting for RAM prices to drop so can start running locally. Been looking at tutorials to run very small models locally shakespeare and tinystories just to learn more.

[jeffburg]

Thanks for bringing this up. As you know from my post on my AI Assistant I made for iOS 5+ and PowerPC Macs, I am deeply into AI development. I am a professional software developer, so my starting point is a bit different, but I would say at this point I would classify myself as AI pilled. So take what I say with a grain of salt. But I feel like DHH getting up in front of 1000 Ruby Developers and ranting about how I am rewriting everything in rust because its faster and better and I don't need to look at the terribly ugly code (reference). I want to be clear, I understand both AI and DHH are controversial and I am not a superfan of all aspects of either of them. With AI particularly, the licensing of code seems dubious. As well energy usage is a problem. But I just can't ignore what it gets for me.

I am now writing toy apps for my favorite retro devices in no time at all. And because of AI I am writing them in pure C (except the UI which is obviously my original love affair, Objective-C). But the important part is that I am writing the highest performance, most reliable, most "sophisticated" code I have ever written in my life. By sophisticated I mean that I add in nice quality of life features all the time I would never write myself on a side project because it's just too much work. The revelation is so great that at this point, I am able to make my iMac G4 (2003) and my iPhone 5 (2012) actually usable in 2026 with all the things I need to do. They all have:

• iCloud Email, Contacts, Calendar Sync
• Youtube video download and playback
• AI Personal Assistant
• Social Media (Mastodon)
• iMessage Messaging
• LINE Messaging
• News (RSS+The Old Reader)
• Web Article Viewing (thanks to the reader proxy from next community actually)
• And more is coming

I write about all of this in my unenshittification series on my blog. But basically, I am unenshittifying my digital life and having a ton of fun doing it. That said, I have not been pushing it very hard online because the retro tech community seems to be pretty averse to AI... and honestly I just don't get much engagement. That does not bother me as I am building things for myself but I really think the retro tech community should try to embrace it (despite obvious concerns)

I think the reason is that for retro software... no one is going to make any money, so there is no incentive to write things. It's pure passion. And in that case having this acceleration and being able to step out of your comfort zone are so critical. Like for me, I am not a C developer and I don't want to become one. But with AI, I can write thousands of lines of C code in a few days... when I look at it, it looks fine. It's so simple, C is such a simple language that I can easily see what it is doing and that it's doing it in a simple way. And I can see its doing memory management correctly (at glance). I am not an expert so I can't tell if it's doing a buffer overflow or something. That part is just determined by the static analyzer and using the app.

But thats the other part about retro software development. The license doesn't matter, it's all free and open source, offered without warranty. So if something breaks or loses data, well, sorry I tried my best but its a free app I wrote in my free time and you should not use a 20+ year old computer in production anyway.

Lastly, is the speed of acceleration. When I started this journey in January, it was pretty slow and painful to write the code with AI. But now, 6-8 months later... honestly, I actually never tell the AI what to do. I tell it my idea, ask it what it thinks and ask it how we can do the most simple and high performance implementation of that thing... and after piling these up after many small small sessions, you have a high performance app that does modern things on ancient computer. And thats because the simple fact is that 99% of modern apps just download JSON files and then display them in a list.... thats Twitter, Instagram, Ebay, WhatsApp, etc. You don't need a supercomputer for this task. It's easy for the CPU. The reason these apps are so slow on our insanely fast modern supercomputers is because they are just full of slop code... not AI slop... purely human created Slop code that existed long before AI existed.

So yeah, again, I apologize for my rant. But I just absolutely love AI for retro software development and have no plans to stop... even if sharing it in our communities is totally taboo. I hope it's not and I hope we can all learn to embrace it together. But I think it's one of the greatest things that could happen to retro computing.

I think we are only 1 year away or so from a total novice from being able to write a modern app on a modern device from a simple prompt. Retro apps are harder to write, so maybe that same milestone is 2 years away on retro apps. But I think we are not far from enabling anyone to make the apps they want for their favorite and most beloved retro devices and I am super excited about it.