Post

This explains so much

Since writing that post, I’ve now come to the view that so much of the software layer at Anthropic is built with the (conscious or unconscious) assumption that eventually the model will do it better than the software.

“Hot-fixes in Claude Code will be RL’ed into the next model, so why be precious about it?“ is perhaps the thought. But yes, I agree with your take as well and think existing expertise continues to be under appreciated as both an enabler and driver of agentic coding decisions. For better and worse.

(BTW, I don’t agree with the ‘the model will eventually handle it,’ approach and wish for native apps. I just observe this to be an Ant POV.)

anthropic.comHow Claude Code is used in practiceNew Anthropic research looking at interactive agentic coding. We evaluate the composition of tasks, human-AI collaboration, and success rates.

Also, this position/culture seems to extend to all UX outside the model, including Ant’s habit of producing some of the consistently worst chart crimes lately: www.anthropic.com/research/cla...

The Sane 🇨🇦❄️

@thesane.bsky.social

If you want people to continue to have a deep suspicion of vibe coding, having the app ‘work’ the way it does is a great way

Raphael Rychetsky

@raphaelfyi.bsky.social

Being responsible for some of the worst Mac apps in the last decade is certainly an achievement

YJ Soon

@yjsoon.com

10 July

Ok so did you see what happened to the ChatGPT app today 😮‍💨 (Spoiler: old one is retired, new one is 1.5GB and on Electron)

Iestyn Lloyd

@iestyn.de

That last line though😆

Iestyn Lloyd

@iestyn.de

That last line though😆

Marc Hummel

@marchummel.com