The Claude Code Desktop app is an affront on software. As developers, we should be offended that they chose to ship something this awful. Rushed out my video because I feel like I'm going insane.

Yeah....

🙃🙃

Everything proves that Mythos is just a hype thing. Otherwise they would have used it to fix it, or maybe the slop is so big that even Mythos said "imma head out". Could've avoided all of this by just open sourcing all of it.

They vibe coded the entire app lmao. I thought claude was supposed to be good at the UI. What reason not to use codex now?

Don’t you think some of these early UX issues are pretty normal for a brand new desktop app from a team that’s moving super fast on the AI side?

No.

@Asymtode712 so the issue is launching too early?

@Asymtode712 It feels like a lack of care

As a senior engineer, I get the frustration the app clearly feels rushed in places. But calling it “an affront to software” is a bit dramatic for a brand new desktop app on day one. Early UX issues, clunky flows, and missing polish are pretty normal when shipping v0.1 of a complex AI-native tool. Give them a few weeks of updates. Most teams improve fast once real feedback rolls in. What specific part felt the worst to you?

Claude Desktop: Built by AI, tested by no one, offended by everyone. 😂

Just 100T more parameters bro. That’s all I need.

hard not to love this guy

I don’t see one claude team member responding to this and assuring that it’s an early version and will relentlessly work on improving the UI/UX/DX If this exact reaction video was for Codex, I’m sure that team would jump to embrace the criticism and feedback Thanks @theo

Respectfully, 90% of the bugs you showed here are things I've never once cared to do. Throwing a fit because cmd+w didn't close the window sort've proves how deep you had to go find issues with it. Just curious, who are you invested in? W Anthropic

Been running Claude Code CLI in production with multiple parallel agents for months. It's genuinely excellent. The desktop app feels like a different company built it. CLI is the real product, the GUI is a rushed storefront bolted on top.

Why dont they just ask Mythos to build it? Just ask it, "Fix Claude Code, Make no mistakes"

Yes, **Theo Browne** (@theo on X, CEO of / T3 Code) has a clear **conflict of interest** when harshly criticizing Anthropic's Claude Code Desktop app. ### Primary conflict: - He is the **founder and full-time CEO** of **T3 Code** ( / a **competing GUI/code editor tool** built around AI coding agents. - T3 Code explicitly supports and wraps **Claude Code** (the CLI version from Anthropic), positioning itself as a polished, open-source alternative or enhancement layer. - In his recent posts (including around the same time as the rant), he openly promotes T3 Code, shares updates about adding Claude support, reassures users that using T3 with Claude won't get them banned, and contrasts it favorably against "vibe-coded" or unreliable tools. - He has stated that T3 Code was created because his team was "unhappy" with existing options (including Anthropic's and others like Cursor) and wanted something better/more trustworthy. Criticizing a competitor's official desktop app while actively developing and promoting your own alternative (that integrates with their backend) creates an obvious incentive to highlight flaws and steer users toward your product. ### Other relevant context: - Theo has a history of public friction with Anthropic: He has called out them "DMCA'ing" his Claude-related forks, cracking down on third-party tools using Claude subscriptions, and restricting access for competitors like xAI (via Cursor). - He positions T3 Code as open-source and community-driven ("we're hiding NOTHING"), which he uses to build trust and differentiate from "closed" or poorly built alternatives. - He is also a YouTuber and investor who frequently discusses AI coding tools, so his content naturally drives visibility (and potential adoption) to T3 Code. That said, his technical criticisms (bugs, poor UX, lack of polish) appear genuine based on the live demo in the video — many developers echo similar frustrations with rushed AI-built software. Theo also criticizes other tools like Cursor in private channels and has praised aspects of competitors when he feels they earn it. In short: **Yes, there is a direct business conflict**. His strong negative take on Anthropic's desktop app should be viewed through the lens that he has a financial and competitive interest in users preferring (or at least considering) T3 Code instead. This doesn't automatically invalidate his points about quality, but it explains the intensity and timing of the rant.

$400B company ships absolute garbage. It’s distasteful that we normalize this.

Love t3code, wanted to use it as the GUI layer for Claude/Codex running inside VMs on but got stuck on remote auth Waiting on -auth-token flag!

the CLI was built by people who use CLIs. the desktop app explains itself.

seriously what the hell are even these extra toggles in CC settings? pretty hilarious

Mythos is failing miserably

I think Anthropic is under enormous pressure with the limited compute that they have.

I haven't watched the video (yet)... but I'm not so offended. I like it better than the prior built in "code". If they continue to improve, fix bugs, and make remote control auto start (and work)... this is the first time I could see non-tui in my workflow for cc.

A 30-min rant over minor inconveniences disguised as "bugs" just to shill T3 is exhausting. The CLI was good enough already even for non coder like me. Anthropic shipping a sloppy desktop app doesn't mean Opus/Mythos are bad; it just means this is low priority compared to models

The sloplification will continue until morale improves.

Make it make sense to me. If you're trying to sell the industry on agents being the 'future of programming,' your own outputs should be the gold standard. Shipping bug-ridden slop completely undermines the pitch.

The pressure to ship before its ready is real, but the reputation cost is higher. Users remember the first impression more than the eventual fix.

I think that's the problem with ML and AI guys trying to improvise being UI/UX devs ..they are not (and no skills file will fix that) Apple and others invest billion into UI/UX research The whole AI and agents world seems to be winging it....

I couldn’t login. Seriously tried for 5min and then just gave up.

Codex was as bad at launch

the cli already solved the hard problem, the desktop wrapper just added latency to it.

I don't think the CLI is bad at all, quite useful in actual fact especially when it comes to keeping an eye on what the agent is doing. Where I am unhappy is that I've seen a noticeable degradation in the quality of output to the point at which it's laughably bad and I'm not sure what to do. Do I cancel my subscription? Is it a temporary glitch?

i mean claude code hasn't really been usable for a good while either right... i'm confused why you're surprised by their antics anymore tbh.

The obsession with forcing AI into bloated desktop apps is wild. The best AI developer tools shouldn't even have a UI, they should just be invisible background agents living in the repo, doing chores like updating docs, while we stay in our own IDEs.

Perplexety is the same

I think they made the app by doing vibe code with Claude, in fact you can see it...

You should see the new Gemini app lmao

They should open source it, fans would help for free.