Most vibe coding platforms have a glaring limitation. You describe what you want, the AI writes the code, and you get a web app. It lives in your browser. It forgets who you are when you clear your cookies. It goes down when your internet connection drops.
I keep coming back to this exact frustration. I do not want another browser tab. I want a tool that feels like it belongs on my computer.
That is the entire premise behind Glaze. Raycast just launched this new platform, and it completely bypasses the browser. You type a prompt. A few minutes later, you have a native Mac application. It has a real dock icon and responds to keyboard shortcuts. Best of all, it works entirely offline.
I have spent some time exploring the beta. There is something profoundly weird about having a custom desktop app materialize just because you asked for it.
Moving beyond the web wrapper
The term "vibe coding" usually implies compromise. You trade performance and deep system integration for speed and ease of use.
Glaze breaks that rule. The apps it generates are not just Chromium wrappers disguised as desktop software. They have deep hooks into macOS.
When you build a utility with Glaze, it can access your local file system. It can live in your menu bar. You can assign global hotkeys to it. It runs locally on your machine, which means it opens instantly.
Imagine building a simple markdown scratchpad that auto-saves to a specific folder on your hard drive. It takes maybe three minutes of chatting with the AI. The app boots in milliseconds. It does exactly what you need and absolutely nothing else. You do not have to navigate a complex IDE or figure out how to package a Swift application. You just ask for it.
The engine under the hood
Raycast did not build its own foundation model for this. Glaze acts as a specialized orchestrator for existing top-tier coding models.
Under the hood, it leans heavily on Claude Code and OpenAI's Codex. These models are already excellent at writing software. The magic of Glaze is how it constrains and directs them.
If you go to a standard chatbot and ask for a Mac app, you get a wall of code and a list of instructions on how to compile it. Glaze hides all of that friction. You provide the prompt. The platform handles the compilation, packaging, and execution.
This tight integration makes the feedback loop incredibly fast. If the app does not look right, you tell the chat interface what to change. The UI updates in real time. It feels less like writing code and more like directing a designer who works at the speed of light.
The community directory and remix culture
You do not have to start from scratch. Sometimes staring at a blank chat box is intimidating.
Raycast included a community directory inside Glaze. People are already sharing the hyper-specific tools they have built. You can browse these apps and install them directly.
But here is what gets me. You can also remix them.
Let's say you find a great pomodoro timer in the directory, but you hate the color scheme. Or maybe you want it to trigger a specific AppleScript when the timer finishes. With normal software, you are stuck submitting a feature request and hoping the developer cares.
With Glaze, you just clone the app and ask the AI to make your changes. You own the software. You can mold it to your exact preferences. This completely flips the traditional relationship between software creators and consumers.
The end of bloated software
We have accepted that most apps are bloated. A modern chat application takes up a gigabyte of memory because it ships with a thousand features you will never use. Product managers have to build for the average user, which means nobody gets exactly what they want.
Glaze points to a different future. You can build single-purpose tools that solve your specific problems.
Consider someone who needs to resize images for a specific blog layout. They can drag a photo in, and the app spits out the exact dimensions they need. They do not need Photoshop. They just need that one function.
This is the real promise of AI-generated software. It lowers the cost of creation so dramatically that disposable, single-use apps become practical. You can build a tool for a project that lasts a week, and then delete it when you are done.
The psychology of owning your tools
When you buy software off the shelf, you adapt your workflow to the tool. Think about how many weird workarounds you have developed over the years just to make Jira or Notion behave the way your brain works. We contort ourselves to fit the software.
Vibe coding flips this dynamic. When you build the tool yourself, the tool adapts to you.
There is a profound psychological shift when you realize you can just change how your software works. I found myself looking at my daily tasks differently. Instead of asking what app I can buy to fix this, I started asking what tiny app I can describe to Glaze to automate this.
It brings back the feeling of early personal computing. In the 1980s, people bought computers and immediately started writing basic scripts to manage their recipes or balance their checkbooks. The machine was a blank canvas. Modern operating systems walled that off behind complex developer environments. Glaze tears down that wall. It makes the computer personal again.
How vibe coding changes the App Store ecosystem
Apple has built an empire on the App Store. They control distribution and they take a cut of the revenue. That model makes sense for massive applications like games or enterprise software.
It makes very little sense for a utility that converts timezones.
Glaze completely side-steps traditional app distribution for these small utilities. If I need a timezone converter, I do not go to the App Store or read reviews. I simply tell Glaze to make it.
This is going to put immense pressure on indie developers who rely on simple utility apps for income. The barrier to entry has dropped to zero. If your entire business model relies on a feature that someone can vibe code in four minutes, you are in trouble.
But I also think this opens up a new market. We might see developers pivoting from building full apps to building complex Glaze templates. Instead of selling a finished product, you sell the starting point.
The reality of debugging in plain English
I want to be clear that this is not magic. It is software, and software breaks.
Glaze is currently in private beta, and you can definitely feel the rough edges. When you ask for complex logic or apps with multiple distinct views, the AI can get confused. You often have to guide it step by step rather than dropping a massive list of requirements all at once.
Sometimes the styling looks weird. Sometimes an app just fails to compile, and you have to hit retry. It is impressive, but it requires patience. You are still debugging, even if you are doing it in plain English.
We need to talk about what debugging actually looks like in this new paradigm.
When a traditional app crashes, you look at error logs or check Stack Overflow.
When a Glaze app does the wrong thing, you have a conversation. You say the save button does not seem to do anything. The AI reads its own code, finds the missing connection, and pushes an update.
This sounds great, but it can be intensely frustrating. Sometimes the AI gets stuck in a loop. It will apologize, claim it fixed the issue, and then produce the exact same broken app. You find yourself trying to trick the model into approaching the problem differently. You are not debugging code. You are debugging an LLM's reasoning process.
It is a completely different skill set. I suspect AI wrangling is going to become a core competency for anyone using a computer in the next few years. You have to learn how to be extremely precise with your language. You have to learn how to isolate variables when describing a problem.
What this means for the rest of macOS
Raycast has slowly been eating the Mac operating system. It started as a simple Spotlight replacement. Then it added clipboard history. Then window management. Then AI chat.
Glaze is their most aggressive move yet. By giving users the ability to create native apps on demand, Raycast is positioning itself as an operating system within an operating system.
You do not need Finder if you can build a custom file browser. You do not need Notes if you can spin up a custom text editor. Raycast is becoming the interface through which you interact with your machine.
Apple is likely watching this very closely. They recently introduced native window management in macOS Sequoia, finally matching a feature Raycast has had for years. I would not be surprised if Apple attempts to build something similar to Glaze into a future version of Xcode or macOS itself. But for now, Raycast has the lead.
Official Links
- Project Page / Demo: https://raycast.com/glaze
Conclusion
I genuinely do not know how far this trend will go. We might hit a wall where AI simply cannot handle the complexity of larger applications. But for the small, annoying problems that plague our daily digital lives, the solution is no longer a search query. The solution is a prompt.
If you use a Mac and you are tired of generic software, you should get on the waitlist. You might find that writing your own tools is easier than complaining about the ones you bought.