If the purported benefits of agentic-coding systems are real, it should have been trivial for Anthropic to implement a native macOS application, and sell it as a benefit, no? The whole reason Electron was being pushed is because it reduces development effort, but AI should be able to nullify it based on Anthropic's own messaging.
AI has to get these native builds perfect for each platform whereas with Electron it's like... write once run anywhere evolves into perfect it once, perfect it everywhere.
Absolutely. But there's the support, documentation and testing aspect and how they're setup.
To their credit, OpenAI acquired a team of experienced macOS developers when they bought Sky [1], the basis for the Mac version of Codex. Apple acquired Workflow from the same team years ago, which became Shortcuts, a core part of macOS's automation system.
Anthropic says 80% of their code is written by Claude; the remaining 20% requires some human expertise, especially with platform-specific features.
A native macOS version of Claude Desktop (and therefore Claude Cowork) would be a huge step forward for Anthropic. Hopefully it's a matter of when, not if.
[1]: https://www.macstories.net/news/sky-acquired-by-openai/
Theoretically AI could help with this by making it easier to support multiple interfaces for the same backend functionality but it would run into the challenge of measuring good taste. A good Mac app has lots of interactive aspects which you notice while using it, and would have to be non-trivial to express in ways which a bot could measure.
Somehow I don’t think sucking up 300+Mb of RAM to display 15Kb of text is anywhere within the realm of “performant”, especially when a native app can do the same in 1/30,000 as much RAM.
And with RAM prices to spike another 60-80% before the end of the year, Electron apps are a downright moronic and utterly brain-dead choice, to say nothing about failing to read not only the entire room, but an entire stadium filled to capacity. People are desperate to stretch RAM as far as they can, and Electron is the worst possible choice for that.
I would eagerly categorize any greenfield project using Electron as being absolutely retarded. The mentally incapacitated definition. Because that’s what any such decision-makers are.
I mean, if you truly want a write-once, work everywhere platform without any kind of a required runtime, there is DotNet.
My favourite that kinda settles this debate is - https://news.ycombinator.com/item?id=48168058
SwiftUI is immature but getting better. There have been tons of AppKit apps with excellent text support. Don't know what this guys problem is.
A native macOS app done correctly is superior to a web app in virtually every way. Claude Desktop won't be a best-in-class app if it remains a web app. Simple as that.
Think of how most any modern computer would grind to a halt if every tab was its own full-fat web browser.
And THAT is the problem with Electron.
It’s not one installed framework providing the foundation for whatever number of apps you want to use, it’s a single electron framework for each and every app.
And that shit adds up, fast.
Electron isn’t like the JVM, which exists only once per computer and which can run dozens of different apps from a singular foundation. Electron is a completely self-contained system that will exist once for every app that is built on top of it. You have twelve Electron apps up and running, you have twelve obscenely bloated instances of Electron causing your computer to run out of physical RAM and thrashing the paging file to hell and back.
RAM doesn’t grow on trees. Prices are set to spike another 40-60% before the end of the year, above and beyond their already insane valuations. Anyone looking to use Electron in a greenfield project in this current environment is totally ignoring the current RAM crunch, and is spitting gratuitously in the faces of their users. And IMHO, doing so is moronic. There is no other way to spin that under current RAM conditions.