With the advent of vibe coding, is there really any industry for people developing apps for blind consumers?

Vibe-coding is not a replacement for careful application development. It might work just fine for small applications where the quality of the code doesn't matter much since the models can keep the whole code within their context windows, however current AI sucks at the engineering part of software engineering, so it doesn't take long for the mess that they write to become unmaintainable. AI-generated code is legacy on arrival. There are plenty of things that AI cannot do all by itself. It's easy to build an Ikea chair, but the methods don't scale to building a skyscraper.

The only problem with AI, at least for me, is that it adds a lot of noise to job applications. In the past having a portfolio of small interesting projects that truly showed one's skills was all it took to find a job in this industry; these days that's no longer a viable option because anyone can vibe-code impressive small things, so the only way to stand out now is to work on something large, which requires commitment and specialization. While I have no problem with specialization, because I've learned enough for the last 29 years to compete with specialists of several areas of software engineering, I don't like to commit to a single large project because that doesn't showcase my versatility.

Ever since I realized (many years ago) that NVDA and VoiceOver are as reliable and robust as they are, I certainly am not willing to pay for assistive software in particular. I was once a JAWS user, but having learned that there are organizations that both know how to keep themselves afloat and care enough about their users to keep their products free to use, equal or better than JAWS, and, in NVDA's case, even open source, I find JAWS to be mostly a scam.

More generally, I'm not willing to engage with subscription models for almost anything, software or otherwise. Utilities—phone, electricity, etc.—are the major exception here, because it's understandable to me that those companies need continuous income to fund the services they provide to me. Almost all other subscriptions are subscriptions purely out of greed or control. I no longer accept not owning the media I purchase, and I no longer accept indefinite funding of an app just so I can use features that are already developed, tested, and shipped. These are just a few examples.

I'm not willing to pay at all for any app that I know to be built largely with a chatbot—don't care what it does or how shiney it looks. First, there are moral issues involved. Second, there are the privacy and security questions. And more basically, why would I pay for someone to sit around while a chatbot makes what they want when I'm perfectly capable of also sitting around and letting a chatbot make what I want if I really wanted to? I'm also not afraid to build my own solutions completely independantly, and I find the challenges and learning experiences involved with that process infinitely more rewarding than prompting a chatbot to fake its way to something that's going to end up of objectively lower quality anyway.

I am willing, even happy, to pay for a piece of software once (once per major release even) if it's accessible, well designed, and genuinely useful in a way that I can't easily reproduce. Keyboard Maestro comes immediately to mind as fitting into this category. I'm even more likely to purchase or even to donate when I find compelling evidence that the owners of the project ban the use of chatbots to produce their code. None of these are new stances I've taken since vibe-coding became a thing, however, and I don't think it's very complicated. It boils down to expecting serious, professional, high-quality work in return for my money.

I need products that are developed by people with the appropriate skillset. Using AI to assist with your work, Vs. doing your work for you are 2 different animals.

Yes, I'm also someone who believes jaws is a scam. Way too expensive for what it is. I can buy a secondhand laptop here for cheaper. A very decent laptop at that. Plus, jaws takes up way too much computer resources. Especially with Windows 11 and other applications are taking up enough computer resources already. I feel sorry for those running that screen reader on low and hardware. That's why I'm glad NVDA exists. NVDA is the complete 360 of a scam, it is the polar opposite and cannot be compared to any other window screen reader at all

I’d pay for that. Wouldn’t anyone?

You know, just using what is before you. If you need to pay for it and have the ability, pay for it, if not adjust your mind to use the free or Vibe version, or make it yourself.
But then who is paying for the AI usage? I think that cost has many hidden aspects. Perhaps even a future generation will be paying for the free app just coded by AI.

Dear Oliver,

What I like about vibe coding is that it gives people a way to build something around a need they have themselves. On a big project, the developer’s ideas and experience come into play right from the start. What problem are we trying to solve? Which part of the system should be responsible for what? When something changes, what else might it affect? Those are engineering decisions I’d want to think through before asking AI to write the code.

Before writing a roadmap, I’d set aside time for research. That’s where I’d ask AI for help first. What do the technologies I’m planning to use give me? When do they give me incomplete information? What small experiment could I run to find out? That way, when I start designing the architecture, I know what assumptions I’m making.

Say we’re building a screen reader. We need to take the accessibility data an app provides, make sense of it, and speak the result. A settings window might have a group of tabs, a list in the selected tab, and a button to apply changes. Each of those is a component, a small piece of the interface that makes up the window.

A component’s name, role, and state are a starting point. In a list, I’d want to know whether moving with the arrow keys also changes the selection. For a button, I need to know what action it offers and whether it’s currently enabled. With tabs, I care about which tab has focus, which one is selected, and which panel is showing. Then there’s how those pieces work together. When I switch tabs, how do I reach the new list? When I go back, can I pick up where I left off?

I’d separate the work of getting the accessibility data, interpreting it, and presenting it to the user. I’d run AX queries in a separate worker. In a data-driven design, I’d define lists, buttons, and tabs individually using a shared set of rules. Each definition would describe how to recognize the component, which states to announce, and how interaction should work. I’d have AI write the shared Swift engine that applies those rules. When the same components turn up in another window, the engine can reuse the behavior it already knows.

That also gives you something to build on later. A user could choose the order in which they hear an item’s name, role, and state. If I add braille support, I can use that same state information to choose a symbol. Quick navigation can use the component’s category, and a button’s help hint can use the key binding that’s active at the time.

I’d think about app updates in much the same way. If a control moves or its AX identifier changes, I’d like to be able to update that in a profile. A workflow with its own conditions might call for a script, and a more specialized operation might need a separate native module. I’d make that decision based on how the app actually behaves.

AI can do a lot here. It can write code and tests, look through AX records for changes after an update, and help figure out where the delay after a keypress comes from. Then I’d try the actual task as a user. Can I reach the right place with both Tab and quick navigation? Is the right item selected? Do I hear what I expect in time to keep working?

As for making money, I think much of what makes a product viable comes down to the engineering decisions that give it a solid foundation. You have to find a problem people run into regularly and are willing to pay to solve. For someone to keep relying on a tool they buy today, it has to keep working as apps and operating systems change. That takes judgment up front and care over time. I also think someone who develops that way of working can bring it to a professional software team. Being able to explain why a feature works the way it does, and knowing where to make a change when circumstances shift, is part of the job.

Humans, for now, will have to look over the code, that will change in the future and I can't wait. I can't wait to see how far we can go with computers and their brains but for now? Vibe coding is great. AI is a tool, and like any tool, it will be upgraded and improved.

@Dominic just so you know mate, it's ah, a h, not r, I know why you thought it was, r and ah sound similar in Australian and British English but yeah its a h.

I'm looking forward to the day AI allow us to break free from all things related to subscriptions. While AI itself is a subscription, at least I'll be using just one while kicking all the others to the curb. if you can't get rid of all subscription models, then the sensible thing is to at least minimize them as best you can.

AI can't make anything, it just steals work people have done, mashes it together, and adds a bunch of unnecessary garbage on top of it. Given how very specialized blindness tech needs to be and that it needs to be made by people who know something about us, I wouldn't touch a vibe coded anything with my longest cane. I want people who know how to program so when something breaks they can fix it.

Using AI to decompile something like Voice Dream Reader so more people can access those features away from a subscription by a terrible company might be worthwhile. But that would still require people who knew how to work with what they got out of it. AI might be useful for some things but on the whole it is causing a lot more problems for us than it's able to solve. And it never will be able to solve anything because that's not what it is or does. Six months from now I highly doubt anyone will be able to just casually churn out something the way people have been.

When have blind people ever been willing to pay for apps? I was forced to pay for screen readers, foreced to pay for OCR software, for decades, over and over again. Often I was told the high price was because I was part of a small market, wiht high development costs. Blind people can function as softweare developers now, more easilly than ever. I use coding agents now, for me it is a return to 1994, I first used a coding agent back then, his name was Wil. I was developing some network system or other, I needed a very small Windows utility. Wil worked in a different building on the same site as me, I called him up and told him what I needed. He made it, got it to me and I tested it. Two itterations later it was done. Wil was good, but I have to say, Claude is better!

What happened with Claude? I'm curious, because for me it is fantastic. I keep all my files on Git and CloudFlare puts them online. I've got databases, an MCP, agents that seem to live and run somewhere only we know. Once I got it set up properly, it is like a dream. I've been more productive in the past month online than the past 20 years!

TO me this is a tired argument. Unless you are programming in assembly, or some other 0s and 1s bare metal level, you are using code previously developed by others: programming language, a compiler, libraries, frameworks. Research of shared ideas and solutions from Overstack, etc. An LLM does make it more convenient and faster to code, an LLM does not have the creative idea. There's always been some of this attitude, the first WYSIWYG editors for HTML, the first drag and drop visual programming, oh how I recall some had such hatred for Visual Basic/C++/C#... This is just the next evolution.
And to the original question about is there a software industry, this is a fear across the software industry currently. And yes, I think the industry is still there. The emphasis is going to migrate to software design and engineering, to creativity. If you can solve a problem for people, they will still buy it. I can tell Claude to build me something, but that does not mean it still doesn't take time, expertise (to actually make it work), the patience, possibly the infrastructure build-out for services. I do think that the quality of products will go up for pay, as easy quick things you can make and don't need to buy. But same as most likely all of us have an oven, most of us don't bake our own bread on a daily basis. There is a place for quality software and services. What that need and demand is, well that's the mystery and secret to discover.

are you gonna use the applevis app or test it now?