Adam, thanks for the inspirational story. I was a HyperCard developer and mourned its passing, though I did make do with FileMaker Pro for many years afterwards.
I’ve been using Grok with a SuperGrok Lite subscription to automate some of my work tasks, with impressive results.
Your obvious enthusiasm for using AI to code apps makes me wonder if it might be possible to re-create the abandoned Clipboard History for modern systems. I’ve tried all the current clipboard managers, but can’t get no satisfaction.
ScanSnap Manager (which I use with an S1500), please. I discovered today I won’t be able to upgrade past macOS 27 if I want to keep using it because there won’t be any more Rosetta.
Exactly. The effort spent in programming is usually about the structure of the program/app and the data structures, as these can affect things like the responsiveness of the app. From comments made elsewhere it seems that asking an AI to do a full app without instructing it about structure is a guaranteed disaster.
Great article. It will be interesting to see the application of AI in more formal environments like aviation. I wrote requirements for various commercial avionics products as my career. I’m retired now but it would be interesting to see how one of these AI systems would handle a requirements spec.
I recently had a conversation with one of my coworkers who is a development director (I’m a technical product manager). His feeling is that with AI coding he can more easily turn a product manager into a developer than a developer into a product manager. I agree with that assessment.
For every 10 devs I’ve known, only 1 or 2 really understood the big picture of the thing we were building from a user perspective. The rest could build working software and solve the technical issues or the scaling issues or what have you but left to their own devices the things they built would be a “UI built by a software dev for other software devs.”
Those 1 or 2, though, understood how to get into the mind of the intended user, understand the problem or opportunity we were trying to address for the users, and bring real insightful input to the table besides just the technical / coding aspects.
So my anecdotal experience is there are a lot more people who - like Adam - are technical “enough” to undersrand a problem and frame it in such a way that it can be a software solution than there are developers who can do the same.
My most recent example with the data team at my company. I vibecoded several reports based on the data I know we have available to use as prototypes for the team to build production reports. They started asking me things like “what aggregations do we need to build” - I didn’t say this to them, but if I need to answer those types of questions why do I need them? I did say, “that’s why I’m coming to you, to help turn these prototypes into reality”
Adam, your “technical but not a coder” identity resonates so thoroughly for me, too. It’s how I became a technical project and later product manager. Being able to translate between marketing/business people and developer-type people has been my niche, too. You in journalism, me on the teams.
with that said, I’ve been vibe coding solutions at work for months now and doing things that weren’t possible to get done. Not because of technical feasibility, but priority and cost. The dashboards and tools I’ve built would have been priority 1,465 on the list and there is no way comparatively expensive and scarce dev resources would be allocated to something that only helps 10 or 100 internal employees. But those tools have unlocked things for us that are worth 50x the minimal cost of the tokens.
My recent personal AI coding experience was trying to export my medication log from AppleHealth. The “export” button includes a list of meds (I’m only on one) but not the log of when it was taken. There are HealthKit APIs that do expose that, and ChatGPT even recommended a couple of apps that could get it for me. But to get more than just a week were paid Apps ($25). All I wanted was my med log to share with my doctor this week.
I asked codex “build me the simplest app possible for me to export my medication log”. I did have to manually install Xcode and login to my personal Apple developer account to be able to sign the app, but otherwise it took under 10 minutes for me to get my full medication log. The app has one button and uses the default share sheet so I can download the resulting CSV.
This article gave me some perspective on my own very positive experience. I have dabbled with software projects since the days of Fortran and 80-col cards but have never taken a computer science class.
A few years ago I developed a web-based PHP app that basically compiles information from many sources into a visitor information display for use in parks. It is in use in a national park. Like many non-professional single-developer projects, mine had some sloppy code and poor choices, but it solved a real problem. I wanted to redesign it with a robust maintainable architecture and good design practices, and I enlisted Claude to help me out. Wow. What a great experience. As you described, it was more like being a project manager working with an expert tutor than being a coder, discussing (with a machine) alternatives and tradeoffs, and occasionally nudging my coding partner back from a non-useful rabbit hole. I probably wrote a dozen lines of PHP. The result was a solid, much better product in a few months that probably never would have happened at all if I tried to do on my own.
But I can’t help wondering how I would feel if I were 25 years old and had just completed 4 or 5 years of computer science classes. AI has turned knowledge and experience into a commodity.
Adam, what a brilliant tour. I shared your love of HyperCard when I discovered it on someone else’s Mac. I bought my first computer, an Atari ST 1040. Then I discovered a plug-in Mac Plus emulator, and that was all I needed to start building stacks, in my case animation for pitching a game I invented, and also a simple Hypertext authoring tool. I mourn its absence, although your article has once again distracted me from my career in film editing with hopes of prompting AI to build possibly useful apps, and even movies. I have seen some astounding narrative movie work from friends made entirely from prompts. My head is spinning.
You documented aphantasia, but most of the article reveals a talent for imagining useful tools and describing what you wanted from them. Not being a psych, it’s possible you’ve misdiagnosed yourself, I sense a visionary at work. Keep up the fine work!
Best as always,
Loren Solomon Miller
NeoTron Cinema Group
If you want to put it in the App Store, yes. Just to run it on your own devices, I don’t think so, but you should confirm that.
Building and distributing apps for Apple platforms is pretty convoluted, and differs across macOS and iOS.
Xcode itself is a free download and you can experiment with all this stuff without paying Apple anything. You can use Codex or Claude to try out your idea for an app for just the cost of the AI.
But… on macOS, there are three categories of apps: locally signed, notarised, and App Store.
You can build and run your own code (or Claude’s code!) using local signing. It can’t use things like iCloud or push notifications, and distributing the built app to other people is cumbersome because each user will need to explicitly enable it to run against Gatekeeper’s wishes.
Notarised apps require the annual Apple developer subscription but enable iCloud, push, etc. and can be distributed however you like to other people without scary warnings. There are still some things they can’t do, like in-app purchases, but they are also not subject to sandboxing which is important for some kinds of apps.
Distributing through the App Store also requires the developer subscription and enables in-app purchases, etc., but are sandboxed and limited in various ways.
On iOS, it’s a bit different.
Without a developer subscription you can use Xcode to install apps from source code on your own phone, but you can’t distribute the built app to anybody else. These apps also can’t use iCloud, push, etc. This would be ok for many tinkerers, except annoyingly the installations expire after a few days and need to be continually rebuilt from Xcode to keep them working!
To permanently install your own apps on your own phone you need the developer subscription. This also lets you submit apps to the App Store and access all the features of iOS like iCloud, push, in-app purchase, etc.
Enjoyed the article. Like many here I too was able to grasp and build some functional things with Hypercard.
For the recognizing a runner and tagging a clip to the person problem: instead of facial ID or bib # OCR, could the bib include a NFC tag? Then when a runner passed a reader it would trigger an action. Disney does this with their “magic bands’“ to animate things when the wearer passes or pauses at a specific point.
The problem with NFC is that the small flexible tags that you could reasonably attach to a piece of clothing have very short ranges. The runner would pretty much have to tap the tag against the reader.
Tags that work at a distance (like EZPass) require a powered transmitter, making them significantly larger and requiring a battery. They’re also not dirt-cheap the way passive tags are.
Magic bands have a battery in them. When the battery dies, all those at-a-distance features stop working and you can only use the features that require tapping a reader (like opening hotel door locks and paying for merchandise).
And Disney, in their infinite wisdom, didn’t make the batteries removable. So once it dies, that’s that.
Adam, There are a number of apps (I still think of them as programs) which run on a PC but are unavailable on a Mac. I recently purchased a LiPo battery charger which has the ability to update the firmware. However, the update firmware only runs on a PC.
I used Claude to teach me how to install Bootcamp on my iMac. And, I really needed Claude. There were many pitfalls along the way which Claude navigated me around. I didn’t really have the courage to ask Claude to read the PC version of the app and convert it to an app for my 2019 Intel iMac but in a year or two, I’m sure I would go that direction.
After I’d gone to all that work, it turns out the firmware in my charger is up to date and there is no update available.
I look forward to gaining the courage to use AI to attempt such a conversion myself.
Sorry, Adam; after more than 30 years reading TidBITs, I’m out. Judging by many of the enthusiastic comments here, I’m in a small minority.
For what it’s worth, I also started with HyperCard in 1993, and dabbled a bit in FileMaker Pro 3 & 4. I don’t see any connection to working with those tools and using a magic 8 ball to generate code.
I’ve been hearing you on various podcasts describing your enthusiasm for AI tools in general, and now vibe-coding specifically, and it’s been disappointing to hear.
Do I think LLMs will end the world? Of course not. But the people behind these tools have repeatedly shown themselves all but incapable of ethical behaviour, committing theft on a global scale that none of us should be prepared to forgive-and-forget simply because there’s convenience to be had or money to be made. Theft is theft, to say nothing of the environmental havoc being wreaked by data centres feeding these tools/toys. Or chatbots that encourage kids to commit suicide or execute school shootings (as Sam Altman bravely apologized for via letter to grieving families of victims in British Columbia recently).
Can’t imagine you’ll care what I think, but this makes TidBITs just one more tech publication that’s drinking deeply of koolaid that I see as poison.
Enjoy it while you can. We’ll all be paying for it when this bubble inevitably collapses on us.
But I will add on a somewhat more serious note that I often benefit greatly from new AI abilities in web browsers and such places. I’m not a programmer, but I like to use AppleScript applications built OMM. A lot of them. A Lot! It’s easier for me to remember a single AppleScript app with a single function that I have created, even if I created it 20 years ago, than it is for me to learn how to navigate multi-purpose apps. Once in a while I run into an old homemade app that was created for Intel which gives me the warning in Tahoe (yes, I finally moved to Tahoe a week ago, now that it’s been hammered into shape some, wasn’t too bad, so far).
My default broad search is Google in Safari on my Mac, so I asked it, “please show me the actions of a shortcut that will take an existing intel applescript application and replace it with a native Apple Silicon version.” I figure “please” is quite superfluous, but I’m used to it.
The answer was rough around the edges, but I would have spent 30, 40 minutes or an hour if I started from scratch, and in about 10 minutes this was doing exactly what I wanted. You have to start by getting the file system objects that you want to update, of course, and Google did steer me the wrong way by suggesting “Ask for files” as the first action rather than the correct “Select files”, but the code it provided was much more streamlined and correct than I would have started with.
So I’m democratized, more or less. It’s hit or miss.
I don’t view LLMs and generative-AI to be any more of a threat to society than other technologies were when they were new.
Printed books, the telegraph, the telephone, radio, television, USENET, email, websites, instant messaging, push notifications, blogs, wikis, social media…just about every major advance in how people obtain information caused some hysteria and, yes, examples of misuse and abuse.
But humans learn and adapt.
And let’s face it: most worries and predictions of doom related to new things become quaint and dated with the passage of time. The world survived the publication of books by Charles Darwin, D.H. Lawrence, and Henry Miller, the broadcasts of Elvis dancing, and students looking things up in Wikipedia.
With you completely, tantramar. I am beyond disappointed in the casual dismissal of all that is dangerous with AI. Adam’s “sure there are problems but I’m having fun” is like toasting marshmallows on a house fire. After subscribing for decades I cannot support TidBITS any longer.
I’d argue that you have it backwards. The use of AI, and consequences from data centres or the ethics of companies creating AI, are short term problems that are simply a history repeat of technology advances.
The real concern is the threat posed to humanity. And it’s not a theoretical risk, even people who build AI (I mean the actual engineers, not the CEO’s) are concerned about the risk.
Electricity and water use of data centres is a short term problem. AI companies run by “tech bros” are not unique in their dubious ethics. The changes coming to ALL of society and the impact to humanity is the real challenge. The solution for that is a long way from claims that “AI is bad”.
I use AI every day. I also discuss with others the existential risk of AI every day. The two are not mutually exclusive.
“Electricity and water use of data centres is a short term problem.” Perhaps in places where that is the correct spelling of “center;” I simply don’t know. But in the US, those areas where proposed data centers appear to be concentrated are seeing drastically higher residential electricity rates, ostensibly to build the infrastructure to supply power to dozens of AI “data” centers. In the meantime, at peak electric rate time, the data centers will burn methane in bank after bank of generators. Better than coal, but not by enough. The excess energy generation will take a toll on the climate at the same time extreme weather, intensified and made more frequent by global warming, is reducing rainfall across North America and western Europe, and exceptional heat waves are helping to fuel forest fires and brush fires on both continents. Ah, but Microsoft says it will use nuclear power to run its data centers: good from a climate change perspective, unknown from a public safety one.
Aside from fears of a Skynet-like AI taking over a nation’s lethal weaponry, the greatest potential downside of AI is the tech bro CEOs who stand at its forefront. They, and their venture capital backers, helped by IPO investors, are fueling a firestorm of silicon pipeline monopolization that is driving up the cost of all other tech drastically. In the short run, that could be ruinous; in the middle-to-long term, whether because AI competition sorts itself out to a small number of winners and a vast cesspool of losers, or the whole thing falls into a black hole, much of the economic value in corporate shares could be wiped away in a matter of months. Simply too much concentration of capital in too small a sector of the economy that represents too much risk.
I am not opposed to people in the beginning and middle of their careers trying to make money. Previous generations shouldn’t pull the ladder up after themselves after experiencing a great deal of opportunity and many advantages during their lifetimes.
In a related vein, I would be very interested in seeing research on how correlated people’s current environmental concerns related to AI usage and data centers are to previous behaviors and attitudes. Fringe suburbs, large motor vehicles, big houses, and long commutes aren’t environmentally friendly either.