Overview
Casey Newton uses this episode to test Glaze, Raycast's new Mac app for "vibe coding," and to ask a simple question: what changes when app-building moves out of the terminal and into a live visual editor. His answer is modest but clear. Making software gets a lot more inviting when you can see it update in real time, and that makes highly personal, small-scale apps feel more plausible than they used to.
Key Takeaways
Glaze seems to lower the barrier for non-technical users by starting with a working Mac app, compiling and installing automatically, and letting the user edit visible parts of the interface while the app is open. Newton says that alone removes a lot of the scaffolding and prompt-writing that tools like Claude Code still require.
The strongest point in the piece is not that AI-built apps are better than standard software. In one case, Newton says the opposite: there is no real reason to build another to-do app when Todoist already exists. What changed is the cost of making something weird, personal, and maybe unnecessary. His Nightwing-themed task app is silly by design, but it made chores more enjoyable for him. That matters because software usually asks users to live with its bad parts. Here, the user can change them.
The more serious example is his Platformer archive app. He had already built a version with Claude Code, but friction kept him from using it often. Glaze let him turn the same idea into something that sits in his dock, surfaces recurring topics and people, and gives him a faster way to search his own reporting. The point is less "AI makes new products" than "AI can turn a useful prototype into a tool you'll actually open."
His unfinished contacts app, Sourcecode, points to a bigger use case: private software shaped around one person's work. Newton wants a local, CRM-like system for sources that can track job changes, notes, and documents. He has not solved the privacy, security, or design issues, but the project shows where this style of app-building may have real value: software nobody else would build, but one person badly wants.
There is also a quiet warning running through the piece. Newton jokes that his comic-book image generation probably raises copyright and guardrail problems. The tools are getting easier before the rules are getting clearer.
Practical Steps
If you want to try this kind of app-making, Newton's experience suggests a few useful rules:
- Start with a problem that annoys you often, even if it seems small. Friction is enough. His archive app came from not wanting to open the terminal every time.
- Build for one user first. The best examples here are personal tools, not attempts at a mass-market product.
- Use AI app-builders for interface-heavy desktop tools, especially when seeing live changes will help you refine the idea.
- Do not treat novelty as proof of value. Newton's to-do app is fun, but he is honest that it does not beat the category leader on utility.
- Watch privacy and security early if your app touches contacts, archives, notes, or source material.
- Expect your first version to be half-baked. The payoff comes from being able to keep changing it as your needs shift.
Notable Quotes
- "It's fun to make things, it's fun to make things."
- "Until recently, software development was too expensive to pursue something so stupid. Not anymore." - Casey Newton
- "The new vibe coding tools promise a world where whatever sucks about the software you use can be changed instantly." - Casey Newton
Full Transcript
This is Platformer Plus. I'm Casey Newton. The following column was created using a synthetic voice clone made by Eleven Labs. In today's episode, vibe coding has escaped the terminal. Adventures with Raycast's new app-making app, Glaze. This is a column about AI. My fiancée works at Anthropic. See my full ethics disclosure at platformer.news.ethics. I began the year with a flurry of vibe coding projects, starting by canceling my Squarespace account and building a new personal website, and proceeding to build a suite of tools for creating daily briefings, a journaling companion, and a platformer archive. The projects were so useful and so easy to create that initially I did not spend much time thinking about how unattractive they were. Aside from the website, gorgeous, my new tools were either accessible through the terminal or Markdown files, functional but fugly. I've continued to use those tools intermittently. Over time, though, their lack of polish came to nag at me, and so I took notice when, in March, Raycast announced Glaze, an all-in-one vibe coding app for the Mac. Where most coding tools are programming generalists, Glaze was built with the express purpose of building and sharing desktop apps. It was not the first to this idea. Wobby, whose founder, Eugenia Quida, I interviewed here last month, does the same thing for mobile apps. But as someone who does most of my work on a laptop, I found Glaze more immediately appealing, particularly given Raycast's track record. Its launcher app is one of my favorite and most used pieces of software. I added my name to the waitlist and forgot about it. Then, at the beginning of June, my invitation showed up in my inbox. Last week, Glaze opened up to all users. It includes a free tier with a one-time bundle of credits that will let you build an app or two. After that, you'll need a pro subscription, $20 a month at launch, which refreshes with 200 credits monthly and lets you buy more if you run out. In my first month, I built three apps that I've been using to accomplish various tasks. All of them involve productivity in one sense or another, though only one has any practical utility that extends beyond what was already available in the App Store. Still, I have found the process of making Mac apps delightful, even when what I have made is extremely silly. And while ultimately my conclusions about the next generation of vibe coding don't extend far beyond, it's fun to make things, it's fun to make things, and that has made me modestly more confident that there is a future for hyper-personalized software, at least in the short term. So with that, here are a few notes on my early efforts to usher in the SaaS-pocalypse. By far the most ridiculous app I have made so far is called Nightwing, after the DC superhero. Recently, in an effort to watch less YouTube, I have begun reading a comic book or two before bed. My original comic book phase had taken place in middle school. I hadn't picked up a new one in 30 years. But something about the profusion of AI-generated slop in all my social feeds had made me desperate to see and pay for some handmade human art. And so a couple months ago, I bought subscriptions to Marvel and DC's respective comic book apps and began digging into what I had missed. One superhero whose stories I found myself enjoying was Nightwing, Batman's original Robin, all grown up and leading a team of superheroes in various world-shaking conflicts. Around the time I got access to Glaze, I had the amazing, terrible idea of creating a Nightwing-themed to-do app. To-do apps are the classic first vibe coding project, and there is almost never any good reason to make another one. I can't imagine anyone meaningfully improving on Todoist. Still, I find myself changing to-do apps roughly every six months for aesthetic reasons and for the relief that comes from starting fresh with a blank slate of projects. Within a few minutes of using Glaze, its advantages over a pure coding app like Claude Code became apparent, at least to a non-technical person like myself. Glaze starts with a working Mac app template, which lets you skip lots of prompting and scaffolding work. It compiles and installs the app automatically. Another big time-saver. It calls image generation models natively, which Claude Code doesn't. And perhaps, most helpfully, you can edit your app while it's open, even drawing a circle around specific elements you want to change to guide the coding agent. You can publish your app to Glaze's store if you want, or browse other people's creations to find inspiration. For the most part, I have found apps in the store pretty boring. Featured suggestions include a PDF reader, a subscription tracker, and a Claude usage monitor. My Nightwing app is defiantly absurd by comparison. At the click of a button, I can generate an image of a Nightwing-esque character doing any task that's on my list. Sending an email, buying groceries, getting an MRI. When I mark it completed, comic book-style boom and pow animations fly across my screen. A widget at the bottom of the app offers a synopsis of a random issue of the Nightwing comic book, while the issue number at the top of the app tells me what day of the year it is. Is any of this remotely necessary? Of course not. Does it represent a copyright violation and an indictment of the safety guardrails of the LLM provider I am using to generate these images? Almost certainly. In the meantime, though, my little app has succeeded where no previous to-do app has to date, making my chores fun. For me. Until recently, software development was too expensive to pursue something so stupid. Not anymore. Moreover, all of these features came about through me typing what I wanted into a box and watching the app update live in front of my eyes. Those live visual updates made vibe coding feel easier and more interactive than it has been so far. After Nightwing, I built a better platformer app. For the first version, I used Claude Code to ingest the publication's entire archive and allow me to run semantic searches over it. This is useful for reminding me what I've said about a particular subject and when. Summarize my recent coverage of Meta's oversight board, for example, is more useful than a Google search, particularly on deadline. But running those searches has required launching the terminal, launching Claude Code, and then invoking the skill associated with this particular project. It's not hard, exactly, but I've used it less than I might have otherwise due to the friction involved. With Glaze, I quickly built a platformer app that now lives in my dock. Like the Claude Code app, it can run queries over the entire archive using my Anthropic API key, but it also extracted the key topics and people I return to over and over again, and I can browse my coverage of them by clicking on the chips that sit on the app's homepage. I also added a feed of my most recent 10 columns to the homepage, along with randomly selected quotes from the past six years. If my Nightwing app is silly, this one is quite serious. It's a cheap, elegant research assistant that no one else would ever have built for me, and I put it together in about a day. It's the sort of app that I can imagine being useful to writers of all sorts, whether for journalists keeping track of beat coverage or fiction writers trying to stay on top of their lore. And it can be adapted to the writer's exact needs, whatever those might be, and however they might change over time. My final project is still half-baked. It's a contacts app that I'm trying to adapt for my journalism called Sourcecode. I've never been great at keeping track of who I talk to and when, and with Sourcecode, I've combined the contacts in my phone with my contacts on LinkedIn to begin to build a more visual graph of my network. What I really want is to create a CRM-like system for updating my network's contact information automatically as their jobs change, a kind of private, locally stored LinkedIn where I can add notes and documents and perhaps run semantic queries as well. Who do I know who used to work at Google? For the moment, I'm thinking through issues related to privacy and security. I'm also not sure whether I even like my current design or whether I want to pay for a CRM service. Still, I feel a sense of excitement and possibility that I never previously felt while updating a contacts manager. So much of the historic experience of using software is having to put up with the things that suck about it. The new vibe coding tools promise a world where whatever sucks about the software you use can be changed instantly. And now, here's Elmar Kianos on what else we're following. China weighs restricting AI. Here's what happened. This month, Chinese authorities met with AI developers to discuss potentially restricting international access to China's most advanced models, Reuters reports. Companies including Alibaba, ByteDance, and Zid.ai were involved in the talks. In the past few years, Chinese AI companies have become known for their open models, LLMs close to the frontier that developers can download and customize. China's Ministry of Commerce is now in discussions with those companies to restrict these open models as well as closed ones. Officials talked about making any leak or theft of proprietary AI tech an offense under China's national security law. Chinese authorities are very worried about Anthropic's cyber-capable Mythos model, Reuters reports. As a result, officials are weighing policies similar to those the Trump administration adopted this month. U.S. officials have asked OpenAI and Anthropic to limit their most powerful models to a small group of approved organizations. And only in the past week lifted export restrictions on Anthropic's Mythos and Fable models. Here's why we're following. Mythos and the Trump administration's strong reaction to it are drawing