Daring Fireball: Claude’s Criminally Bad Electron Mac App Is an Inside Job
By John Gruber
Archive
The Talk Show
Dithering
Projects
Contact
Colophon
Feeds / Social
--><br>Sponsorship
Day One — The journal you actually keep. Start with a chat, end with a journal entry. ⭐ 4.8 (400k)
Claude’s Criminally Bad Electron Mac App Is an Inside Job
Friday, 3 July 2026
Anthropic released the first version of the Claude “desktop” app for MacOS in October 2024 — an Electron clunker that did not impress UI designers. When it came out, I wrote:
ChatGPT’s native Mac app, on the other hand, is a truly<br>native Mac app. It looks like a Mac app and feels like a Mac app<br>because it really is a Mac app. I’ve liked it ever since it<br>launched back in May, and it keeps getting better. And<br>I keep using it more and more as my go-to resource for answering<br>questions.
I asked Claude, “What is the best way to engineer a native Mac<br>app? What frameworks and developer tools should one use if the<br>goal is a great Mac experience?” Claude’s answer<br>started by positing it as a decision between SwiftUI and AppKit.<br>Perhaps Anthropic’s Mac engineers should have asked Claude this<br>same question before they built this turd of an Electron app.
In March of this year, linking to Anthropic’s announcement that Claude Code and Claude Cowork can take control of your Mac to accomplish agentic tasks, I returned to the same question:
The Claude Mac client itself remains a lazy Electron clunker. If<br>Claude Code is so good I don’t get why they don’t prove it by<br>using it to make an even halfway decent native Mac app.
I’m not the only one who has pondered this. Drew Breunig wrote “Why is Claude an Electron App?” in February this year:
On the surface, this ability should render Electron’s benefits<br>obsolete! Rather than write one web app and ship it to each<br>platform, we should write one spec and test suite and use coding<br>agents to ship native code to each platform. If this ability is<br>real and adopted, users get snappy, performant, native apps from<br>small, focused teams serving a broad market.
But we’re still leaning on Electron. Even Anthropic, one of the<br>leaders in AI coding tools, who keeps publishing flashy agentic<br>coding achievements, still uses Electron in the Claude desktop<br>app. And it’s a slow, buggy, and bloated app.
So why are we still using Electron and not embracing the<br>agent-powered, spec driven development future?
For one thing, coding agents are really good at the first 90% of<br>dev. But that last bit — nailing down all the edge cases and<br>continuing support once it meets the real world — remains hard,<br>tedious, and requires plenty of agent hand-holding. [...]
For now, Electron still makes sense. Coding agents are amazing.<br>But the last mile of dev and the support surface area remains a<br>real concern.
I’m with Breunig up until the point where he accepts coding agents struggling with the final 10 percent as a justification for choosing Electron to create a Mac app. Plenty of people — individuals and teams alike — are using Claude Code to create terrific new native Mac apps. Just among my friends, Glenn Fleishman, Lex Friedman, and Jason Snellman, have all in recent months used not just AI coding assistants in general, but Claude Code specifically, to create genuinely native Mac apps that meet their own personal high standards for Mac-assedness, forged through decades of literally professional Mac snobbery. A comprehensive catalog of Mac-assed apps made with the assistance of Claude Code, would, I suspect, be remarkably long.
The struggle with the last 10 percent is unrelated to AI coding. It’s the nature of all software engineering. There’s a well-known adage that Wikipedia names the “Ninety-Ninety Rule”, attributed to Tom Cargill of Bell Labs:
The first 90 percent of the code accounts for the first 90 percent<br>of the development time. The remaining 10 percent of the code<br>accounts for the other 90 percent of the development time.
Cargill’s mathematically humorous formulation resonates because it not only explains why the final 10 percent consumes half the time, but also why software projects tend to take twice as long as expected. This universal truth holds whether the code is human-written, AI-generated, or a mix of both.
Breunig gets closer to the truth in a postscript, linking to the Hacker News thread discussing his post. The top-rated comment in the HN thread is from Boris Cherny, who works at Anthropic on the Claude Code team. Cherny wrote:
Boris from the Claude Code team here.
Some of the engineers working on the app worked on Electron back<br>in the day, so preferred building non-natively. It’s also a nice<br>way to share code so we’re guaranteed that features across web<br>and desktop have the same look and feel. Finally, Claude is<br>great at it.
That said, engineering is all about tradeoffs and this may change<br>in the future!
I would rephrase the guarantee that “features across web and desktop have the same look and feel” as...