Couple of weeks back, I built a teleprompter app for my Mac. A transparent overlay that scrolls my markdown script while I record videos, with timing markers so I can control the pace at specific points. I’d looked for teleprompter apps before, but none of them did quite what I wanted. This one does, because I built it for myself.
This week, I built a Substack Notes scheduler. Not a Chrome plugin that hijacks the browser, but a menu bar app that sits quietly on my Mac and lets me queue up notes and post them from the terminal. Tools like this exist, but they all charge a monthly fee for something that felt simple.
Both of these took about a day to build. Both do things that the paid alternatives don’t. If you’re selling downloadable software, the kind of app someone buys for ten bucks, the customers who would have bought yours are starting to build their own instead. The market for simple desktop tools has always been built on an 80/20 compromise.
That missing 20% takes two forms: sometimes the tool doesn’t quite fit the way you work, sometimes what you’re paying for doesn’t really need a subscription in the first place. My teleprompter was the first type, the existing ones were close but none of them did what I needed. My scheduler was the second, there’s no reason scheduling a Substack note costs money every month. Either way, you used to accept the compromise because getting “exactly what I want” would cost thousands of dollars or months of work. It stops making sense when the cost drops to an afternoon.
With Claude Code or Cursor or Codex or whatever you prefer, you can describe what you want, iterate for a few hours, and have a real tool that does exactly what you need. It runs on your machine, it costs nothing to operate, and it doesn’t need support or distribution or a backend. Markets are evaporating, because customers aren’t switching to a rival product, they’re just leaving.
Not everything is in danger
Not everything is equally exposed. It helps to think about what actually protects you.
Network effects, even small ones. It can be as simple as two people needing to use the same tool at the same time: a CRM, a shared notebook, a project board. The moment your product requires collaboration, building your own version stops being a weekend project. Someone has to keep the database running, handle the auth, make sure it’s up at 3am.
Complexity. I use a product called Cap for screen recording, and it handles a level of technical depth that would take months to build properly even with AI assistance. The more technically demanding the problem, the less likely someone is to build their own. And once a tool needs a backend or handles data for multiple people, that’s not downloadable software anymore, that’s infrastructure, and infrastructure isn’t going anywhere.
Taste. The reason my teleprompter has timing markers, keyboard shortcuts, and multi-desktop support isn’t because I’m a developer. I’m not. It’s because I’ve spent years as a product manager thinking about how tools should work, what edge cases matter, and what makes the difference between something you tolerate and something you reach for. Designers and engineers with strong product sense bring the same kind of informed perspective.
The people building their own tools right now aren’t random consumers. They’re PMs, designers, technical founders who already understood what good looked like but previously couldn’t or wouldn’t code it. And those were probably your most demanding customers, the ones filing feature requests, the ones who noticed every rough edge. They’re the first to leave, precisely because they cared the most, and when they go they take the feedback loop with them. The people who made your product better are the ones walking out the door.
That leaves two groups still buying: people who can appreciate good design but can’t create it themselves, and people who genuinely don’t care as long as it works. The second group is probably your most reliable customer base, they’ll pay three bucks a month forever rather than spend an afternoon building something. The first group is more interesting. They might still prefer your product over something they could prompt into existence, because they value your point of view. They’re paying for your opinion about how the tool should work, and that’s something you can’t generate from a prompt. For them, the 80/20 compromise still holds, because your 80% is better than their 100% would be.
The future is here, but it’s not evenly distributed
It would be the understatement of the century to say that I’m excited by this. I go to sleep thinking about it, wake up with new ideas, and it’s made every idea I have feel realisable. My LinkedIn feed, Substack Notes, and YouTube suggestions are all filled with people who are equally excitable.
The enthusiasm is bouncing off the walls in my echo chamber, but most people don’t care that this is possible. They have better things to do than read threads about vibe-coding or follow along with AI tool releases. The person who would have bought a teleprompter app for ten bucks probably isn’t spending their afternoons prompting Claude Code to build one.
The people who are building their own tools are a specific group: people who already knew what they wanted and now have the means to make it. That’s a growing circle, but it’s not mainstream, and it probably won’t be for a while. You need to enjoy the process, or at least care enough about the outcome to invest the time.
That said, there’s a separate threat that doesn’t require anyone to build anything. Platforms absorb features. Substack could add note scheduling tomorrow and most of the reason my scheduler exists would disappear. Riverside already has a teleprompter, though not quite what I want. Claude is now generating charts and visualisations right inside the conversation, which means one less reason to buy a dedicated charting tool. There’s a thin layer of apps whose entire market gets swallowed by the AI tools themselves, not because anyone built a competitor, but because the capability just showed up inside something they were already using.
Though this cuts both ways. LinkedIn already has scheduled posting, but I’d still love to add LinkedIn scheduling to my little tool because then I have one place to manage everything. Buffer is a popular product even though most of the platforms it supports already have their own scheduling. Sometimes the value isn’t the feature itself, it’s having one single place to handle it.
The floor is rising
If you’re selling simple, local software today, you probably still have a few years. The circle of people who can and will build their own tools is growing, but it’s not mainstream yet.
But the floor is rising. The baseline of what people can build for themselves gets higher every few months. If your only defence is “it works and it’s convenient,” that’s the 80/20 compromise, and the compromise is getting worse for you every month.
The tools that survive this will be the ones that have something AI can’t easily generate from a prompt: a genuine point of view about how the tool should work, real technical complexity under the hood, or the kind of collaborative infrastructure that requires someone to keep the lights on. Ideally more than one of those.
Anyone can now build the tool of their dreams and leave it on their desktop, never to share it with anyone. But if you want to sell it, if you have a point of view worth paying for, the problem becomes distribution. Getting it in front of people. That might be the strongest moat of all, because somewhere out there is another brilliant person with a slightly different opinion who built something just as good. They just left it on their desktop.