Update:
I’d written a pages long post about why GenAI is an enabler, which you can still read below. The process of working through my thoughts, however, brought to mind serious questions about privilege, access, and intent, which brought me back to being someone unsettled by the presence of AI.
Am I still going to use it? Yes. Because when I ask it questions like “why does this bitwise AND mask work?” I get a very thorough explanation that brings me closer to the kind of deeper understanding of the mechanics of how this all works in the hardware. This is widely known and available in countless books and learning materials and there’s nothing saying I could not learn that way. But here is this tool can be a personalized tutor. I argue this is fundamentally different than simply asking a robot to make something for me so I can watch a movie.
Perhaps what’s fundamentally different in this situation is that I’m driven by my desire to slow down and learn and I’m bringing years of experience and peripheral broad knowledge to the transaction. I’ve coded for years and know things like memory management exist and how it functions at a very high level—but I want to go deeper.
When I wrote
…do this on my own, with classes and books and human teachers.
That triggered a thought. All those things require money, resources, access. GenAI is free for the time being, but will it be? I’m only using it for building a simple app, but what about uses that affect more of humanity? Ultimately AI must be considered from this perspective of access and equity.
These are questions I don’t want to derail my original post with, but I had to put this preamble out there.
O.G. Post
Skip to the obvious question:
Why?
I asked myself that many times. I already use Ulysses with markdown XL to write my books on a modern Mac. BBEdit and Alpha exist for System 7, and are phenomenal text editors — Alpha is even extensible and probably could be made to parse markdown, so why bother making another? And the probably the most pressing question: is using AI for this cheating, exploitative, or needlessly derivative?
What’s Old is New
When this technology was new, I didn’t have the experience or the budget to justify paying for THINK C. Ten years later, after taking C and C++ classes, learning a little bit of assembly, I eagerly bought an educational version of CodeWarrior on CD.
But I never wrote an app with CodeWarrior. I may never have gotten much past installing it and staring at the blank IDE wondering what to do next. Ultimately, I’m a visual designer who like to code — enough to be dangerous, but not a seasoned computer scientist. Life in general and other academic pursuits pushed it to the bottom of the queue. I was not training to be programmer per se, and I have lots of hobbies already.
Yet I never forgot making choose your own adventures in AppleSoft Basic on my //c. Working with computers has been a part of my life since I was in single digits, which is why I still love making user interfaces delightful in Vue.js.
Even if I’m not particularly skilled as a programmer, what I enjoy is working with a computer to make something happen — I love the process.
What’s Slow is New
Why not write in Swift for modern macs, you say? Modern APIs would save me dealing with a lower level language like C, fussing with RAM at a lower level and parse text. But in fact that’s what I want to learn. Also, you can’t use swift to my knowledge to compile for an ancient target CPU like one in 40 year old Macs — and that is my primary goal: a very specific use case of writing in markdownon my restored 1989 SE/30. I think of it like a writer resorting to a manual or electric typewriter. It’s about the experience and aesthetic of the compact Mac design.
So the natural path seemed to be is to learn C and build the app myself since it doesn’t exist the way I want it. It’s easy to forget how the philosophy behind the Mac OS made application building more accessible which is still apparent when coding with Xcode today. It’s powerful, and they go to great lengths to make it accessible. This is just that same mindset as it stood in the late 80s.
I bought the books. I sat down to code and diligently worked through some of the tutorials in the Macintosh Primer. It’s a great book that steps you through several core concepts of the Macintosh ToolBox, explains itself well.
There’s a tension here that exists in everyday life: how quickly do you need this done versus how well or how elegantly and bunch of other factors. In some contexts speed is the primary difference between outright success or failure: Running races. Launching Rockets. Server response times.
And though I was making progress with the primer, and reading Inside Macintosh and the ANSI C manual, I had so many questions. I wanted clarification of the code.
I thought: what about using generative AI to answer these questions and dig deeper?
It’s Dangerous Out There…
To start, I asked Claude to “get me started on a markdown app for System 7.1”
It decided to create the whole thing in one go—three minutes of later it handed me 5K lines of code, commented, done.
That alone triggered an existential crisis. Why bother? This AI can do this in minutes.
But! It did not compile immediately and I was confronted with bugs I didn’t understand. No problem — Claude provided explanations, but also had to fix it for me because I didn’t know what illegal pointer arithmetic was. After an hour of blind bug fixing it finally compiled.
It didn’t function as expected. Copy and paste were broken, and it immediately felt completely unreliable as a text editor. And for some reason it decided that we needed two panes to the window. I don’t.
I shut down the emulator and called it a day. In that moment there was no deeper understanding on my part to really fix these problems, but if I wasn’t careful I’d be up all night just trying to figure out what and how to ask Claude how to fix it all.
Back to the races
AI is so fast at something like this the goal appeared to be to about velocity again. Eschew the hype and learn to code on my own, with classes and books and human teachers. Dedicate myself to this.
Yes — that is the path many do and have taken. It On one hand, I’m in no hurry, so why not do that? There is no product deadline, no investors, no expectations other than my own.
But I also don’t necessarily have the time and resources to spend years learning this for a hobby project. It’s a key difference.
I’m not building an app to sell. But even if that were the case, GenAI could be used to “get it done quickly” or GenAI could be used to git gud, providing faster ways for me to get unstuck and still get it done myself.
Enabling technology has done for humans forever: made some process easier or faster. Case in point: without the internet, this process would most certainly take years of work and study, consultation with humans that had been down this path before, trial and error. I’d at least need a class or a dedicated mentor.
GenAI can remove that barrier. It can be asked questions, provide context and answers, and even if it is inevitably wrong at times, still gets me closer to truth through dialogue.
malloc confusing 25 years ago. Mentor? I barely know her!
My second prompt attempt was something like this:
I’d like to start again and move more incrementally, building I might build this application alone: start small, define an MVP and code it, then build in features and iterate.
Let’s begin by setting up a one window app with three menus using a resource fork. We’ll be using ResEdit and Think C 6. All this app can do at first is launch, and quit. Then we can step through an understand this boilerplate code.
That changed the dynamic completely. Now instead of waiting for it to finish and having a cup of coffee while it spins, I’m was having a dialogue, switching between typing in THINK C and getting help from the AI.
I had it walk me through what each line and each reserved word was doing. If I got stuck, didn’t understand something, or simply wanted a deeper explanation into why the code worked or how it relates to the hardware, I asked.
The result is process that’s much more like a mentor and not a diligent servant, and allowing me to stay deeply involved in the process. The AI is more than just some magical thing that writes code while go have lunch. I’m learning the syntax, understanding the intent and algorithms, how the code works and why. I’m stumbling through bugs, my own typos, and the process is slooooowwwww — but I’ll emerge with a greater understanding.
N of 1
I don’t have to care if the app is marketable. It won’t be for sale. I’ll give away on gitHub or the Macintosh Repository I’ll likely be the only user anyway. I get to apply 25 years of UI/UX experience to the project and do all the things I like about that process. I’ll get to know and love every line of code — all while avoiding dead ends and overwhelming frustration that might stop me from beginning. AI will make what might take months into weeks, though faster isn’t the goal.
I draw on an iPad because it has undo. In that sense this technology is a great enabler. Digital media makes drawing more fluid, more forgiving, easier to explore. While often technology is employed forvelocity—get things done more quickly, more efficiently, for the sake of maximizing profits or production— I retain a technique similar to real pen and ink, and don’t need speed.
To that end, I could use a typewriter or hand-write my books longhand. I don’t have to print my books on an offset press. But in the end it’s all about the experience, for myself and in the case of my resulting books, for the reader, the “user”. Less about velocity and more about continuity. I can type well and fast enough to get my thoughts out better than I could write it. A manual typewriter would have a satisfying clack of the keys, but the lack of deletion is not my friend (though I try hard not to edit in the moment, I don’t need stacks of paper on the floor, either — messy again, see?)
Hold Fast to Dreams
Hold fast to dreams
For if dreams die
Life is a broken-winged bird
That cannot fly.Hold fast to dreams
For when dreams go
Life is a barren field
Frozen with snow.
These words by Langston Hughes have stuck with me ever since I was made to memorize them in high school. Reluctantly putting aside the valid concerns about AI and copyright, agency, privilege, and the environment — I worry about all these things — learning with a generative AI is perhaps one of the best case scenarios. A relationship where it’s not a crutch but a learning partner; you don’t hand it all your decisions, you leverage its vast breath and depth to augment your own knowledge. It has infinite patience (though its patience is limited by your credits…) and because this codebase is so well documented, it’s a wonderful resource.
I have a childhood fondness for these machines. Sitting down at one for me brings a sense of real joy. I made little games on my //c in AppleSoft Basic, barely understanding the mechanics of what I was doing, but I was very happy doing it.
When the Mac came along, I did more doodling and drawing — some shapes in Logo in PASCAL, but never a full on application. Still — somehow I always knew the Mac and its philosophy towards application structure would be fun and accessible.
Now 40 years later, I finally get to try again. It might be poorly written, incredibly inefficient algorithmically, and used by only me, but it’ll be something I created… and that’s what matters.