Claude's New Cursor Killer Just Dropped - CourtneyR.dev's 2nd Brain

Claude's New Cursor Killer Just Dropped ![rw-book-cover](https://img.youtube.com/vi/WkHdkwDQJ5o/maxresdefault.jpg) 🔗 **Source:** [Claude's New Cursor Killer Just Dropped](https://www.youtube.com/watch?v=WkHdkwDQJ5o) > [!note]+ Document Notes > tags: #ai-coding #developer-tools #desktop-apps #anthropic #claude #cursor #agentic-coding > author: Theo - t3․gg > source: YouTube > reference: https://www.youtube.com/watch?v=WkHdkwDQJ5o > date created: 2025-01-03 12:00 > > # 🏷️ Title: Claude's new Cursor killer just dropped >  ## 🧰 Action Inbox - [H1] Try Claude Code desktop app vs alternatives like Cursor, T3 Code — comprehensive testing needed #ai-coding - [H2] Evaluate GUI vs CLI for agentic coding workflows — screenshot paste, text copying benefits #developer-experience - [H3] Compare open source options (MCP, T3 Code) vs proprietary solutions — licensing and customization #open-source ## 🧭 Table of Contents - [[#📝 Summary]] - [[#📖 Main Content]] - [[#🗒️ Notes]] - [[#📚 References]] - [[#🗣️ Paraphrase]] - [[#📰 Newsletter]] - [[#✅ Action Items]] - [[#🧪 Hypotheses]] - [[#📊 Methodology]] - [[#❓ Questions]] - [[#🔮 Future Research]] - [[#💡 Implications]] - [[#🧠 Learning Styles Assessment]] - [[#📊 Accessibility Assessment]] - [[#🌍 DEIB Assessment]] - [[#✅ Logical Validity]] - [[#🛡️ DARVO Analysis]] - [[#🎯 Conclusion]] ### 🔗 References - [Claude Code Desktop App](https://claude.ai) - [T3 Code](https://github.com/t3dotgg/t3-code) - [MCP (Model Context Protocol)](https://modelcontextprotocol.io) ## 📝 Summary - Anthropic released Claude Code desktop app, integrated into main Claude app alongside chat and co-work features - Significant improvement over CLI but still plagued with UI bugs, poor UX decisions, and missing basic features - Competing tools like Cursor, T3 Code, and MCP offer better user experience and functionality - Key issues: broken hotkeys, permission bypasses not working, poor multi-window handling, forced project modifications - Bottom line: Despite being from a multi-billion dollar company, the app feels rushed and unpolished compared to alternatives ## 📖 Main Content Anthropic finally released their desktop GUI for Claude Code, marking a departure from their notoriously slow CLI interface. The new app integrates Claude Chat, Claude Co-work, and Claude Code into a single desktop application, promising better performance and user experience for agentic coding workflows. However, extensive testing reveals numerous fundamental issues that make the app frustrating to use. Basic functionality like hotkey bindings, permission settings, and multi-window support are broken or inconsistent. The app forces users to modify their projects (adding to .gitignore, creating Claude-specific directories) unlike competitors that work without project modifications. The review compares Claude Code to established alternatives like Cursor (which uses the open-source MCP CLI under the hood) and newer options like T3 Code. These competitors offer better reliability, more features, and in some cases open-source flexibility. The contrast is particularly stark given Anthropic's resources versus smaller teams producing more polished products. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/WkHdkwDQJ5o?t=0) — [H1] Introduction to Claude Code Desktop App - Key idea: Anthropic released desktop GUI version integrated into main Claude app - Evidence: "They stuffed this in the traditional Claude app as well as co-work" #### ⏱️ [01:45](https://youtu.be/WkHdkwDQJ5o?t=105) — [H2] Why GUIs Matter Over CLI - Key idea: Desktop interfaces offer better UX for screenshot paste, text copying, formatting - Evidence: "Screenshot paste... I can see the screenshot that I just pasted" #### ⏱️ [03:30](https://youtu.be/WkHdkwDQJ5o?t=210) — [H2] CLI vs Desktop Text Handling Issues - Key idea: CLI text copying adds unwanted newlines and formatting issues - Evidence: "It has a bunch of new lines. It doesn't need weird things inserted" #### ⏱️ [04:45](https://youtu.be/WkHdkwDQJ5o?t=285) — [H3] First Major Bug - Thread Freezing - Key idea: Agent runs randomly stop and freeze threads with poor visual feedback - Evidence: "Randomly stopped and then froze the thread... icon is still here, which is terrible UX" #### ⏱️ [06:15](https://youtu.be/WkHdkwDQJ5o?t=375) — [H2] Window Resizing Breaks Layout - Key idea: Basic window operations cause UI elements to misalign - Evidence: "I resized the window... it resized everything in weird ways" #### ⏱️ [07:20](https://youtu.be/WkHdkwDQJ5o?t=440) — [H2] File Opening Functionality Problems - Key idea: File operations don't work as expected, files shown but not accessible - Evidence: "The files here, but it can't see the file. So, it's just an empty thing" #### ⏱️ [08:30](https://youtu.be/WkHdkwDQJ5o?t=510) — [H1] MCP Open Source Advantage - Key idea: MCP CLI and app server are fully open source, enabling custom UIs - Evidence: "Codex CLI, which powers it is... the Codex app server... lets you build your own UI" #### ⏱️ [10:45](https://youtu.be/WkHdkwDQJ5o?t=645) — [H3] Split View Feature - Key idea: Multi-window support exists but has focus and hotkey issues - Evidence: "You can hold down command and drag it to do a split view" #### ⏱️ [11:30](https://youtu.be/WkHdkwDQJ5o?t=690) — [H2] Hotkey System Completely Broken - Key idea: Keyboard shortcuts don't respect window focus and behave inconsistently - Evidence: "Window you're focused in does not determine which window the hotkeys work for" #### ⏱️ [14:15](https://youtu.be/WkHdkwDQJ5o?t=855) — [H2] Project Selection UX Failure - Key idea: Can't easily create new chats for existing projects, missing from recent list - Evidence: "I can't like go to T3 chat and make a new chat in T3 chat because it's not recent enough" #### ⏱️ [16:00](https://youtu.be/WkHdkwDQJ5o?t=960) — [H1] Comparison with Cursor and T3 Code - Key idea: Competitors offer superior project management and UI polish - Evidence: "I hover over Shu, I click new, and now I'm in a new chat for Shu" #### ⏱️ [18:45](https://youtu.be/WkHdkwDQJ5o?t=1125) — [H2] Forced Project Modifications - Key idea: Claude Code requires adding files to gitignore and project structure changes - Evidence: "You have to add things to your git ignore... just to use cloud code" #### ⏱️ [22:30](https://youtu.be/WkHdkwDQJ5o?t=1350) — [H2] Performance Improvements Confirmed - Key idea: Desktop app uses significantly less memory than CLI version - Evidence: "Claude desktop app itself is using significantly less memory than before" #### ⏱️ [25:15](https://youtu.be/WkHdkwDQJ5o?t=1515) — [H3] Settings UI Confusion - Key idea: Settings buttons appear project-specific but affect global configuration - Evidence: "Settings button clearly looks like it is associated with the T3.gg project... Nope, it's not" #### ⏱️ [28:00](https://youtu.be/WkHdkwDQJ5o?t=1680) — [H1] Final Verdict and Industry Impact - Key idea: Despite flaws, will become popular due to ease of access over quality - Evidence: "Will still become the second or third most popular option simply because you sign into it the way you do the normal Cloud app" ## 📚 References - [Claude Desktop App](https://claude.ai) - [MCP GitHub Repository](https://github.com/modelcontextprotocol) - [T3 Code](https://github.com/t3dotgg/t3-code) - [Cursor](https://cursor.sh) ## 🗣️ Paraphrase - Anthropic's desktop GUI is a step up from their terrible CLI but still riddled with basic functionality bugs - Open source alternatives like MCP provide better foundation for building AI coding tools than proprietary solutions - Market success often comes from convenience rather than quality, which may benefit Claude Code despite its flaws ## 📰 Newsletter **New AI Coding Desktop Apps: What WordPress Developers Need to Know** The AI coding assistant landscape is rapidly evolving with new desktop applications from major players like Anthropic's Claude Code. While these tools promise to streamline development workflows, our testing reveals significant quality gaps between different options. For WordPress developers, the key considerations are: - **Project Integration**: Some tools require modifying your project structure, while others work seamlessly with existing setups - **Open Source Options**: Tools built on open protocols like MCP offer more flexibility and community contributions - **Performance**: Desktop GUIs generally offer better user experience than CLI tools for complex workflows **Key Takeaway**: Don't just choose the most popular option - test thoroughly to find tools that actually enhance your WordPress development workflow. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer experience: Evaluate AI coding assistants for WordPress development workflows and create comparison guide - Education: Workshop on integrating AI coding tools with WordPress development best practices - Community management: Survey WordPress developers on AI coding tool preferences and pain points #### WordPress - Core: Investigate AI-assisted code review for WordPress core contributions - Documentation: Create guidelines for using AI coding assistants while maintaining WordPress coding standards - Community: Develop best practices for AI tool usage in WordPress plugin and theme development ## 🧪 Hypotheses - GUI interfaces for AI coding will become dominant over CLI despite initial quality issues - Open source AI coding infrastructure will enable better innovation than proprietary solutions - Market adoption will prioritize convenience and integration over feature completeness - Developer tool quality varies inversely with company size due to different development processes ### 📊 Methodology - Comparative usability testing across multiple AI coding desktop applications - Performance benchmarking of memory usage and response times - Feature gap analysis between proprietary and open source solutions ### ❓ Questions - Will open source AI coding tools eventually surpass proprietary offerings in market share? - How much will convenience factor outweigh quality in developer tool adoption? - Can large tech companies match the agility of smaller teams in developer tool innovation? ### 🔮 Future Research - Longitudinal study of AI coding tool adoption patterns among different developer communities - Analysis of contribution patterns to open source AI coding infrastructure - Impact assessment of AI coding tools on code quality and development velocity ### 💡 Implications - Practice: Developer tool selection should prioritize testing over brand recognition - Policy: Open source standards may be critical for avoiding vendor lock-in with AI tools - Security: Desktop AI coding apps require careful evaluation of data handling and privacy - Education: Developers need training on evaluating AI coding assistants effectively - Funding: Investment in open source AI coding infrastructure may yield better returns than proprietary solutions ### 🧠 Learning Styles Assessment Reading level estimate: Grade 10-12 #### 👀 Visual learners - Present: Multiple screen recordings showing UI comparisons and bug demonstrations - Improve: Side-by-side comparison charts, feature matrices #### 🔊 Aural learners - Present: Detailed verbal walkthrough with commentary and reactions - Improve: Structured audio summary highlighting key findings #### 🗣️ Verbal learners - Present: Clear explanations of technical concepts and analogies - Improve: Glossary of AI coding terms, comparison frameworks #### 🧩 Kinesthetic learners - Present: Step-by-step demonstrations of using different tools - Improve: Hands-on evaluation checklist, testing methodology guide #### 🧠 Logical learners - Present: Systematic comparison of features and performance metrics - Improve: Decision matrix for tool selection, ROI analysis framework #### 🧑‍🤝‍🧑 Social learners - Present: Community feedback and developer reactions in comments - Improve: Developer survey results, community polls on preferences #### 🧘‍♀️ Solitary learners - Present: Detailed individual testing and personal evaluation process - Improve: Self-assessment rubric for tool evaluation, reflection questions #### 🌿 Naturalistic learners - Present: Evolution of developer tools ecosystem and natural selection metaphors - Improve: Ecosystem analysis, tool lifecycle comparisons ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure and descriptive content - Improve: Add alt text descriptions for UI screenshots, ARIA labels for interactive elements #### Keyboard navigation - Present: Focus on keyboard shortcuts and hotkey functionality issues - Improve: Include keyboard-only navigation assessment for each tool #### Contrast and readability - Present: High contrast video content and clear text - Improve: Ensure screenshot annotations meet contrast requirements #### Text resizing and reflow - Present: Responsive video player and text content - Improve: Test tool interfaces at various zoom levels #### Media and captions - Present: Video content with transcript provided - Improve: Add timestamps to transcript, chapter markers #### Forms and interactions - Present: Clear descriptions of interactive elements and their issues - Improve: Include accessibility evaluation of tool interfaces ### 🌍 DEIB Assessment #### Cultural breadth - Present: Focus on developer tools used globally - Improve: Consider different regional preferences and constraints #### Inclusive language - Present: Professional technical language throughout - Improve: Define technical jargon, avoid assuming prior knowledge #### Neurodiversity support - Present: Clear structure with timestamped sections - Improve: Provide executive summary, reduce cognitive load in explanations #### Representation - Present: Focus on technical merit over personal characteristics - Improve: Include diverse developer perspectives and use cases #### Psychological safety - Present: Critical but constructive evaluation approach - Improve: Balance criticism with acknowledgment of complexity in software development ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Systematic testing methodology leads to clear conclusions about tool quality - Inductive strength: Multiple specific bugs and issues support general conclusion about software quality - Causal reasoning: Links UI/UX problems to underlying development and QA processes - Analogical strength: Effective comparisons between similar tools and market dynamics #### Fallacies - Ad hominem: Some personal attacks on Anthropic's competence rather than focusing purely on technical issues - Appeal to popularity: Acknowledges but doesn't endorse the fallacy that popular tools are necessarily better #### Tone of voice - Overall tone: Highly critical but technically grounded, passionate about software quality, occasionally inflammatory but supported by evidence ### 🛡️ DARVO Analysis - Deny: No denial patterns observed - acknowledges both strengths and weaknesses - Attack: Strong criticism of Anthropic but focused on technical merit rather than personal attacks - Reverse victim/offender: No reversal patterns - maintains clear stance as user/reviewer - Emotional cues and rhetoric: High emotion but grounded in technical frustration rather than manipulation ## 🎯 Conclusion Claude Code's desktop app represents progress from Anthropic's problematic CLI, but falls far short of competing solutions despite coming from a well-resourced company. The extensive bugs, poor UX decisions, and forced project modifications make it inferior to alternatives like Cursor, T3 Code, and other MCP-based tools. However, its integration with existing Claude accounts may drive adoption regardless of quality issues. For developers, this highlights the importance of thorough testing over brand recognition when selecting AI coding assistants, and suggests that open source foundations may be more reliable for building quality developer tools than proprietary approaches. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad Purpose: Stage CEE units that become Literature notes. Split level: --- ## Full Transcript [00:00] CLI are great, but they're not the best [00:01] way to do agentic coding. That's why [00:03] Anthropic finally put out an updated [00:05] version of the Cloud Code desktop app, [00:07] and it looks great. Wait, that's Codeex. [00:10] Let me switch over to the Cloud app [00:12] quick. Oh, that's T3 Code. Also looks [00:16] very similar. At least you can use the [00:18] Cloud models here, though, right? No [00:20] worry. We'll switch over to the Cloud [00:22] app now. Oh, wait. That's the new cursor [00:25] glass app. Ah, [00:28] okay. For real though, this is the new [00:31] Claude Code desktop app. It isn't a [00:34] dedicated app unlike many of the others. [00:36] They stuffed this in the traditional [00:37] Claude app as well as co-work. So, you [00:39] have the Claude chat, Claude Co-work, [00:42] and Claude Code all in the same app. And [00:44] it is a huge improvement from the CLI. I [00:47] bet it probably even uses less RAM, [00:48] though I haven't checked that yet. [00:50] There's a lot to love here, and I am [00:52] very thankful Anthropic's finally woken [00:53] up to the fact that a slow, laggy CLI [00:56] might not be the best way to do real [00:58] agent coding at scale, especially when [01:00] you're working on multiple things at [01:01] once. Anthropic is really hyped about [01:04] this release. It seems like they've been [01:05] working on it for a long time. I know [01:07] I've been seeing teasers about it for at [01:08] least a few months now. They have been [01:11] putting a ton of effort into baking this [01:13] new desktop app. So much so, they [01:15] actually tweeted about it from the [01:16] official Claude account, which almost [01:18] never happens. It seems like there's a [01:19] lot of hype around this. So, I took some [01:21] time to really play with it myself. And [01:24] I have thoughts. I obviously have a lot [01:26] to say about desktop vibe coding apps as [01:29] somebody who has built one myself. And I [01:31] will share all of my thoughts and more [01:33] right after a quick break for today's [01:34] sponsor. Do you know what Bun Clerk and [01:36] Nvidia all have in common? They all use [01:38] today's sponsor, Code Rabbit. These guys [01:39] get Code Review. Code Rabbit's an AI [01:41] code review platform used by pretty much [01:43] everybody nowadays. like companies like [01:45] Bun, Clerk, Life 360, Ashb, and of [01:48] course, Nvidia, even Jensen loves Code [01:51] Rabbit. There's a reason why. From the [01:52] sequence diagrams to the CLI and IDE [01:55] integrations to the quality of the code [01:56] review that you get, you just feel more [01:59] confident when you ship once you set up [02:00] Code Rabbit. If you haven't tried AI [02:02] code review yet, I don't care what type [02:05] of work you do. You really need to try [02:07] it. It's how a lot of my team started [02:09] using AI in their workflow. So if your [02:11] company has an AI budget and you and [02:12] your team haven't spent it yet, you [02:13] really need to give something like this [02:15] a shot, it changes how code review [02:17] works. To be very clear, those 2 million [02:19] repos running code rabbit right now are [02:21] not projects that are removing the human [02:23] from the code review. That's not the [02:25] point of AI code review. The point of AI [02:26] code review is to let the humans focus [02:28] on the things that actually matter like [02:30] what does the code do? What are the [02:31] flows? Is this achieving the goal? Code [02:33] rabbit can help with those things, but [02:35] where it really shines is catching all [02:37] of the small mistakes that take up so [02:38] much of your brain. I cannot tell you [02:40] how liberating it was the first time we [02:42] set up Code Rabbit to realize I don't [02:43] have to look for these small syntax [02:45] mistakes anymore and I can focus on what [02:47] matters, the big picture. Does the code [02:48] do what we need it to do? Ship less bugs [02:50] and waste less engineering time at [02:52] soyv.link/code rabbit. So let's dive [02:54] into the cloud code desktop app. The [02:56] first thing I want to do before we go [02:57] too deep is justify using these types of [03:00] guies. I have been amazed at how many [03:02] people have talked [ __ ] on graphical [03:03] user interfaces for their agent coding [03:06] without actually trying them. It seems [03:08] like developers just feel like the [03:10] terminal is the end all beall for [03:12] development. If you can use cloud code [03:13] in the CLI, why would you ever go use [03:14] the guey for it? There's a lot of [03:16] reasons. Here's a simple one. Screenshot [03:19] paste. And now I can see the screenshot [03:22] that I just pasted. I could ask and see [03:25] it here in the history and click the [03:26] image and probably even save it. Oh, you [03:28] can only copy it cuz their app sucks. [03:30] But you can get an actual interface for [03:33] things that you're copy pasting. Or [03:34] let's say I want to copy what it just [03:36] responded with. I can select the text [03:39] and copy it. And if I paste this here, [03:41] I'll turn on word rap so it's a little [03:42] easier to see. All the formatting is [03:44] correct. This is the exact same thread [03:47] in the cloud code. And if I try to copy [03:50] this text from here instead, you'll see [03:54] the problem. It has a bunch of new [03:55] lines. It doesn't need weird things [03:57] inserted. It's not a good experience. [04:00] This has bothered me for so long. Even [04:03] just the word wrap breaking in weird [04:05] ways like this because the space got [04:07] wrapped over. It's just bad. You can fix [04:10] it. I'm not saying any of these things [04:12] are end all be all. I'm just saying they [04:14] kind of suck and they are very [04:16] inconvenient. Or here where I pasted the [04:18] image and it attached the image to the [04:21] tool call prior for some reason and not [04:23] to the message I sent. Although that one [04:26] is actually a bug in the desktop app [04:28] because for some reason when you paste [04:30] an image it sends it as a separate [04:32] message. before the message that you [04:34] send because they don't even know how to [04:36] use their SDK. And here is where the [04:38] roasting begins because this should be [04:41] good and it is better than the CLI, but [04:43] it is better than the CLI simply because [04:45] the CLI is such a trash piece of [04:48] software that anything is better than [04:50] it. Not that all CLIs are bad, just that [04:52] their CLI is particularly atrocious. So, [04:55] what are some of the issues I'm seeing [04:57] that make me so upset? Well, the first [04:59] one is that the first time I ran this, [05:01] it did this agent run when I was just [05:03] asking it to go find performance issues [05:05] and randomly stopped and then froze the [05:08] thread. Could this have just been cla [05:10] behind the scenes? Maybe, maybe not. The [05:13] icon is still here, which is terrible UX [05:14] cuz this looks like it's still going, [05:16] but it's not. For some reason, they just [05:18] have to stuff that logo there so you [05:19] know whose butthole it is that you're [05:21] using. Let's ask the same question again [05:24] quick. Also, no copy buttons anywhere, [05:26] which is terrible. So, manually select [05:29] that. Paste. Oh god, that that those [05:32] updates were bad. Yeah, this UI is kind [05:34] of [ __ ] Allow cla to run. Yeah, I [05:38] pretty sure I already put it in bypass [05:39] mode. I don't know why it forgot that. [05:41] It already made a branch. I'm just [05:43] asking it to find performance issues. [05:44] Why is it making changes? Oh, because I [05:47] approved something, so it's editing the [05:49] permissions. That's great. Oh, that's [05:51] really funny. I resized the window so [05:52] you could see better and it resized [05:54] everything in weird ways causing ah god [05:58] if mythos is so good why is this app so [06:01] bad it missed a few things here that [06:03] aren't true cuz it's opus which is a [06:06] good model but is not a super smart [06:09] model and often misses things well at [06:11] least it responded properly and then [06:13] made a branch and PR will give me a [06:16] button to merge that which is [ __ ] [06:18] cringe as hell apparently file Open file [06:22] lets you open a file. Oh, you said it [06:24] does nothing. It did something. It put [06:26] it there. I don't know why that's [06:27] useful. But if I'm in a thread and I do [06:29] it, will it still work? Okay, that's [06:30] hilarious. The files here, but it can't [06:33] see the file. So, it's just an empty [06:35] thing. You can click the package. JSON [06:37] name to copy that to clipboard. God, [06:39] they this it just doesn't feel like [06:41] they're trying. It really doesn't. And [06:43] not only does Codex do a much better job [06:46] at all of these types of things, Codex [06:48] is also built on top of the Codeex CLI's [06:51] app server, which is fully open source. [06:54] Again, Codex on GitHub, the whole [06:56] package, not the UI. So, the Codeex [06:59] desktop app itself is not open source [07:01] sadly, but the Codex CLI, which powers [07:04] it is. And the Codex app server, which [07:06] is provided by the Codeex CLI, lets you [07:09] build your own UI like this. That's why [07:11] so many of these tools support codecs [07:13] out of the box because you don't even [07:14] need to build a harness. You just plug [07:15] into theirs. And from my experience, you [07:17] can point most agents at this file, at [07:20] these docs, and they will build [07:21] something that works, which is really [07:22] cool. And if you're worried about the [07:24] licensing, don't be. It's Apache, too. [07:26] You're very safe to use this. Not to [07:28] just glaze OpenAI, I'm still pissed at [07:30] them for keeping this closed source. I'm [07:32] so pissed at them that it was a huge [07:33] inspiration for making T3 code. If the [07:35] Codeex app was open source like the CLI [07:37] is, I probably wouldn't have made T3 [07:39] code. But it's closed source, so I had [07:41] to make something else. Thankfully, [07:43] these core pieces that I needed to build [07:45] it are open source and are well [07:46] supported. Where with cloud code, I [07:48] don't even know if I'm allowed. The gap [07:50] here is insane. It is absurd. It is [07:51] frustrating. I am tired. I am upset. I [07:54] wish Anthropic would at least put effort [07:56] in if they're not going to let me do it [07:58] for them. I think you reasonably have [07:59] two options. Either you put out the [08:02] things I need to build what I want, or [08:03] you build something so good that I don't [08:04] want to. And Anthropic is failing [08:07] absurdly at both. But there are some [08:09] things that I found cool. Like if you [08:12] rightclick any other thread, you can [08:14] click open and split view. I think you [08:15] also Yeah, you can hold down command and [08:17] drag it to do a split view like that. [08:19] And if you ignore the awful page layout [08:21] shift when that happens. Yeah, it's [08:25] pretty cool. They also have a hotkey to [08:27] open up the terminal, which I believe is [08:29] command uh sorry, control backte. So I'm [08:32] currently in the right tab here. I'm in [08:36] the right chat. So obviously when I hit [08:38] the hotkey of control tilda, it's going [08:40] to open up the terminal in that one. Oh, [08:42] it opened it in the left one even though [08:43] I was focused on the right one. And I [08:45] can't tab out because it uses the tab as [08:48] an input in the terminal. So the window [08:50] you're focused in does not determine [08:52] which window the hotkeys work for, which [08:53] is uh great. Also, you can't hit that X [08:56] button because the grabber for the [08:58] terminal is touching it. [09:02] I'm not convinced that anybody who said [09:04] they've been using this app for weeks [09:06] actually used it. I hit like five bugs [09:09] like this before I even made a change [09:11] with this app. I feel like I'm going [09:13] mad. I thought that anthropic models [09:15] were supposed to be good at UI. What [09:16] happened? There are some features that [09:18] are promising like you can set up remote [09:20] control so other boxes on your network [09:22] can be used to do things. So if I want [09:25] to like set up another Mac on my [09:27] network, I can connect to that and run [09:29] sessions there from here. It'd be a lot [09:31] cooler if you could do it from mobile, [09:33] but their mobile implementation is [09:34] really focused on cloud code for web, [09:36] which is an entirely different product, [09:38] by the way. Cloud code for web is their [09:40] cloud option here, which uses cloud code [09:43] in the web without the ability to really [09:45] run the code. [09:48] Also, you have no insight here as to [09:50] what project your new thread is other [09:53] than this little button where you pick [09:55] which project. And it doesn't even have [09:57] the ones on the side here. So, I can't [09:59] like go to T3 chat and make a new chat [10:02] in T3 chat because it's not recent [10:04] enough to be here. So, I have to go [10:06] manually find the folder to make a new [10:08] T3 chat chat. What the [ __ ] Every other [10:11] one of these apps has all of this stuff, [10:14] right? If I hop into Codeex, I hover [10:17] over Shu, I click new, and now I'm in a [10:19] new chat for Shu. I can click this drop [10:20] down and pick any of the others in the [10:22] order they are in on the sidebar. I [10:25] switched to round now in a different [10:26] project. Like, are you kidding? Going to [10:28] do a risky thing and install the nightly [10:31] for T3 code that Julius just put up so [10:33] that I can show you guys how we're [10:34] cooking some of these things. And now we [10:36] have the nightly built. And not only do [10:39] we have our projects, we actually go out [10:40] of our way to grab the favicon if it's [10:42] placed in any reasonable location that [10:44] we can find, so it's easier to see which [10:46] projects are which, but we also have a [10:48] new add project flow that I'm obsessed [10:51] with where you can just start typing and [10:53] very easily select whatever project you [10:56] want. Even better than the built-in Mac [10:58] OS file picker. It's so much better. And [11:00] yes, I know that the YouTube video [11:02] compression is not doing me any favors [11:04] with the grays. will probably change the [11:06] colors a bit just for my compression or [11:07] maybe adding some theming, but man, it's [11:09] so much nicer to add projects with a UI [11:12] that makes sense for it. Speaking of UI [11:14] that makes sense, this UI just makes [11:16] much more sense. I don't want to just [11:18] sit here and tell you guys about how [11:19] much better T3 code is than everything [11:21] else. But if you haven't tried it, do [11:23] it. We'll do a more in-depth comparison [11:24] of all of the guies for Agentic Coding [11:27] in the near future because there are a [11:29] lot of them. Like a lot of them. So, if [11:34] that's interesting to you, let me know [11:36] in the comments and I'll do a breakdown [11:37] comparison showing all of the different [11:39] ones and where their strengths and [11:40] weaknesses are. Obviously, I have my [11:42] preferences, but that doesn't mean I'm [11:44] not trying everything else and seeing if [11:45] there's good ideas to take. So, let me [11:47] steal, man. What are the good ideas [11:50] here? Work trees builtin is cool. Work [11:52] trees built in and checked by default. [11:54] Less cool, but I get it. I think it's a [11:57] big commitment for them to do that. That [11:59] is kind of cool. You can add multiple [12:01] folders as context before starting a [12:03] thread so it knows that these are [12:05] folders it can and should access and [12:06] check into. That's a nice idea. The idea [12:08] of being able to pull in context from [12:10] your system without changing the [12:12] directory you're operating in. That's [12:14] nice. Normally I just copy paste a path, [12:15] but this is cool. The coolest thing in [12:17] my opinion though is what I was showing [12:19] off earlier with the tiling. The ability [12:21] to show multiple chats at once in a view [12:25] like this. Very cool. If only [12:27] commandwing didn't put you in weird [12:30] states. Like I have this here. I'll put [12:32] this one to the side. And now I'll [12:34] command W the one on the right. And it [12:36] does what you expect. It just closes [12:37] that. But if I command W the one on the [12:39] left, it puts us in the new chat view. [12:43] Anthropic. [12:45] You know what? I'm going to be a little [12:47] less harsh on them because I know how [12:49] this happens. AI is really good at [12:51] building the happy path. If you describe [12:54] to an agent a user experience and a [12:57] flow, even if it has to touch multiple [12:59] different things, these models, these [13:01] harnesses, these tools have gotten good [13:03] enough to do the thing. But as soon as [13:06] you hit an edge case, like I'm sure that [13:09] at some point somebody gave the request [13:11] to the model of make sure command W will [13:14] close the thread that's currently [13:15] focused. And they did that as long as [13:18] it's not the one on the left. If I [13:19] command W the one on the right, it goes [13:22] away. But if I have two open, I command [13:23] W the one on the left, it does this view [13:25] because it's an edge case. And models do [13:28] not find edge cases. Users find edge [13:30] cases and then have to describe them [13:32] with the model in order to hopefully get [13:34] the model to behave. There is a message [13:36] in chat I want to go in on a bit here. [13:38] The way the window splits is not going [13:40] to gain or lose the users. Bigs 360, you [13:44] couldn't be more correct with this. And [13:46] this is the problem. And this is the [13:48] reason that I am being as loud as I am [13:50] about anthropic. Too many of y'all just [13:52] accept the [ __ ] slop. The level of [13:55] slop that is being shipped by Anthropic [13:57] is unfathomable and y'all just bear and [14:00] grin it. You just use it anyways. [14:03] Despite the fact that there are dozens [14:05] of better options that have more [14:06] features, more functionality, more [14:08] reliability, many of which are open [14:10] source, too. You just don't care. And [14:13] now that Anthropic is directly competing [14:16] with what we have with T3 code, I will [14:18] say I am a bit scared because no one's [14:20] going to choose this interface. Of the [14:22] current options that you have to pick [14:24] from, the Cloud Code desktop app is [14:27] close to the bottom, if not the bottom. [14:29] But it will still become the second or [14:31] third most popular option simply because [14:34] you sign into it the way you do the [14:36] normal Cloud app and it just works. It's [14:38] two less clicks to sign in than other [14:40] options, but then 15 more clicks to do [14:42] anything. And nobody is paying for their [14:44] cla so they have access to this [14:47] interface. They are paying for their cla [14:49] because they like the models, they like [14:51] the inference, they like the answers it [14:53] gives to their questions or the [14:54] solutions it generates for their code [14:55] problems. And they're willing to pay a [14:56] subscription in order to have more [14:58] access. But instead of Anthropic just [15:01] embracing the community and making it [15:03] easier for us to build things like this, [15:05] they put themselves in such a shitty [15:07] state that even they caused themselves [15:08] bugs. Again, I pasted a screenshot here [15:11] and it linked it to a message prior. [15:14] Let's try this again. I'll do a [15:16] different thread here. Going to paste. [15:18] You know what? I'll do this two [15:19] different ways. Also, there's no easy [15:20] way to delete that. I have to select it [15:22] and then press the delete key. There's [15:24] no button for it. I can rightclick [15:26] delete. That's cool. But I'm going to [15:28] switch to the CLI to show you how this [15:30] is supposed to behave. Pasting the image [15:33] with controlV, not commandV, because the [15:35] terminal can't have images pasted into [15:37] it traditionally. And it puts it in line [15:39] in the chat as image number one. I ask, [15:41] what is this? And it puts image one [15:44] underneath as an attachment with image [15:45] one in the copy. I I can't believe they [15:47] actually shipped this like this. It's so [15:49] bad. But it's below where it belongs. [15:53] Oh, it didn't update in the UI here. I [15:56] have to manually refresh it to get the [15:58] update in the UI. And now image one is [16:00] in the text here. Ready for this? [16:02] Pasting the same image here. Asking it [16:04] to double check. It's no longer in the [16:06] chat. I have to re-resoom here for it to [16:08] update. Waiting until it's done first. [16:11] Oh, if you resume before it's done, it [16:14] doesn't even show you that. That's [16:15] great. Yep. So, the first time it [16:18] attached as I showed before, and now I [16:20] said double check, and I put the image [16:22] before. Are you kidding? like it's just [16:24] so little taste. There are nicities in [16:27] the UI like uh let me find the post like [16:30] someone pointed out that the hover [16:32] transitions and the menu hovers are [16:34] better than average. Here's one that [16:36] pissed me off. You have to choose where [16:38] to store git work trees for isolated [16:39] coding sessions and the default is [16:41] includes. [16:43] So you now have to go update your git [16:45] ignore to ignore this or handle it [16:47] properly when you use their workree [16:49] implementation with the defaults which [16:50] is great. Like as far as I know, cloud [16:53] code is the only option where you have [16:55] to modify your project to use it [16:57] properly. You have to add things to your [17:00] git ignore. You have to get it to stop [17:02] dumping [ __ ] in the cloud folder or [17:04] something else just to use cloud code. [17:07] Cursor has their cursor rule stuff, I [17:09] know, but you don't have to do that to [17:11] use it. You have to do this with claude. [17:14] And thankfully, Curser and many others [17:16] are now embracing the standards like the [17:17] agent MD and the agents directory skills [17:20] and all of that. Anthropic still makes [17:22] you put everything in quad files, quad [17:23] MD, the quad directory, all of that. One [17:26] more, and this is petty, but it pissed [17:27] me off. Having a more button with one [17:30] additional option, and then customize [17:32] sidebar. Are you ready to see what [17:34] customize sidebar has in it? I actually [17:37] can't believe how pathetic this all is. [17:42] How' they take so much time and it still [17:43] feels rushed out? Like how? Reminder [17:46] that the Codeex app is actually quite [17:48] good. Has no problems like this? Like it [17:53] might not be as pretty, but it reliably [17:56] works. Things are where they should be. [17:58] The new buttons behave how they should. [18:00] You can drag and drop to reorder things. [18:02] You can pin stuff in a way that makes [18:04] sense. They have their automation stuff, [18:05] which is fine. These all work. Like it [18:07] it just works. It is a little laggy at [18:10] times, but it works. I honestly think [18:13] the biggest change with the new Claude [18:15] Code desktop app is that we finally have [18:17] an official interface from Anthropic [18:19] that won't [ __ ] your computer up [18:20] performance-wise. Actually, let's see [18:22] how right I am about that. I am scared, [18:24] admittedly. Let's see what's wasting my [18:26] CPU now. Now, the virtual machine [18:29] service for Claude is using 2 and a half [18:31] gigs of RAM. Yeah, the Claude desktop [18:33] app itself is using significantly less [18:34] memory than before, which is a progress. [18:38] It's using less than the CLI is, as [18:40] funny as that is. So, yeah, they fixed [18:42] the performance issues, but what's left [18:44] is everything else kind of sucking. [18:46] Like, I like that. Um, what? [18:50] I clicked settings here. Wait. Oh god, [18:54] that's terrible UI. I'm not crazy that [18:56] this settings button clearly looks like [18:58] it is associated with the T3.gg project [19:01] here, right? Like, I folded that. These [19:03] two are part of the same, right? Nope, [19:07] it's not. This is for all things. This [19:10] inline T3.gg settings button is not for [19:13] T3.gg. This is for all projects. [19:18] That was really funny. [19:21] When I change the one at the top, the [19:24] drop-down disappears because again, the [19:27] menu button is literally in that row. So [19:30] if I turn off image studio 2 or T3 chat, [19:32] which are below, the dropown stays open. [19:35] But when I change the top one, the drop [19:37] down disappears because it's literally [19:39] rendered in line with that menu item. [19:41] I'm trying to swear less in my videos [19:42] and Anthropic is trying to make sure I [19:44] can't do that because god damn I people [19:47] in chat are saying they must not have [19:50] QA. Like actually though, it's insane. [19:53] I'm sorry, Theo. It physically hurts to [19:55] watch the UX in that piece of hot [19:56] steaming [ __ ] Lol. I'm getting [19:58] distracted. Do they have no QA? And then [20:00] all of the laughs. Yeah, this feels like [20:03] they're testing the limits of how much [20:05] you can ship a UI that was like [ __ ] out [20:08] with a single prompt. This is [20:10] unbelievable. It is so bad. Oh, at least [20:13] they have forking now where I can fork a [20:15] chat. That's cool. Does it show me here [20:18] if this is a work tree or not, though? [20:21] How do I know what branch this is on? I [20:24] don't think I can. Okay, pantic hersel. [20:27] And yeah, it made a new work tree and [20:30] doesn't give me any insight into that [20:31] anywhere. Of course not. Why Why would I [20:33] need to know that I'm in a work tree? [20:35] That definitely not important. One thing [20:37] they did here that's nice is they made [20:38] it easier to turn off Opus 461 mil, [20:41] which is from my experience a much [20:43] dumber version of Opus 4.6. Previously, [20:45] you had to go edit config files yourself [20:47] to turn this off. I actually switched to [20:49] Opus 4.5 in my local cloud code, which [20:51] has helped even more. Oh, look. Even [20:53] though I put on bypass permissions [20:54] always, it still asks again. Great. [20:57] Yeah, chat asked, "It's on accept edits [20:59] again. Does it really not save the [21:00] preference?" Yep. Oh, if you use voice [21:03] and have multiple chats. Oh, I got to [21:05] test this. What else should I add to [21:07] this project? [21:10] [gasps and sighs] [21:11] Are you kidding? [21:14] Does stop stop the thread or does it [21:16] stop the transcription? I know it's [21:17] going to stop the thread, so I'm not [21:18] going to click it. Oh my god. You know, [21:21] they vibe coded the multi-tab thing [21:24] because god damn. And just for [21:26] reference, here's a dev build of T3 Code [21:30] where I added multi-tab [21:33] with scrollable. And if you open up a [21:36] terminal in one, it opens up in the one [21:37] you have selected. This was two prompts. [21:41] Two prompts. No excuses. I'm done. I did [21:45] actually come in here excited to try [21:47] this out, but I just every feature feels [21:50] so beyond halfbaked. Okay, here's the [21:52] sidebar diff view. Clicking this doesn't [21:55] close it after opening it. That opens [21:57] more things. That closes it. Oh, that [21:59] popping's bad. Why are these not showing [22:02] changes? What happened there? Does it [22:04] not render if it's just additions? Like, [22:06] why are half of those not rendering? [22:09] That Oh god. Can we take a moment to [22:12] appreciate this icon and this icon? [22:16] Sidebarification. [22:17] That one's really funny. It leaves that [22:19] button in even when it doesn't have [22:20] enough space to open. So, right now, the [22:21] second internal sidebar is open. [22:24] We have to go far enough for it. And for [22:26] some reason, it didn't unfold all of the [22:28] things in this diff. Oh, it's not doing [22:30] it again. See, these two are shown as [22:32] opened, but they're not opened. I had to [22:35] close open it to show me. I just They [22:38] say they've been using this for all of [22:39] their work. Either they're lying or [22:41] their work isn't very serious because [22:42] this is just like, are you kidding? And [22:45] like T3 code has had its fair share of [22:48] embarrassing things. I'm not going to [22:50] pretend otherwise. We are two people [22:52] building in the open doing our best to [22:54] get it working to the best of our [22:55] abilities. But god damn, this is just [22:58] slop. I'll make the same request I did [23:00] here, but I have to manually select this [23:01] text because I don't have a click to [23:03] copy. You know, if it was open source, I [23:05] would legitimately contribute fixes for [23:06] a bunch of these things cuz it's not [23:08] that hard to fix most of these things. [23:10] Reminder, we have a one-click copy. We [23:12] have resizing that doesn't break the [23:15] whole app. And once it makes these [23:16] changes, I'll show you that we also have [23:18] a diff view that actually works. Sorry. [23:20] Wait, do they not have an open in your [23:21] editor of choice button? Oh, there's an [23:23] update. I can relaunch to apply. Want to [23:26] see how many of those bugs have been [23:27] fixed? I wonder if they just set a cloud [23:29] code instance to listen to me as I [ __ ] [23:32] and then prompt it. Okay, so it's [23:34] updating, but I don't see any proof. Oh, [23:36] there it is all the way in the back. [23:37] Cool. Does voice still input in both [23:39] text boxes? Yes, it does. Wonderful. How [23:43] about the terminal? It open in the right [23:44] one. Say, I bish this on Twitter. Nope. [23:48] None of the bugs are fixed. [23:50] Guys, there's two reasons as a lab to [23:54] build an app like this. Reason one is [23:57] that you want to lock users in because [23:59] your goal is to try and get users to be [24:01] part of your ecosystem so they will [24:04] continue using your stuff so that if a [24:07] new model comes out for somebody else, [24:08] they'll just stay in your app and not [24:10] even use it. That's reason one is the [24:12] lock inside. That's why they're being so [24:13] shitty about the terms of service [24:15] because they don't want people to move [24:16] to another app where it's easy to switch [24:17] off of their stuff if their models don't [24:18] keep up. That's reason one. Reason two [24:21] is because you're trying to showcase how [24:22] powerful your models and the [24:24] technologies are. That's why everybody [24:26] on the cloud code team constantly brags [24:27] about how they use cloud code for all of [24:29] the changes they make. This is a [ __ ] [24:31] showcase. And I hope I pray that people [24:35] who actually code seriously are as [24:38] insulted with the lack of basic QA [24:40] etiquette that exists here enough to use [24:43] anything else. Anything else. Again, two [24:47] people working on T3 Code, the open- [24:49] source app, were able to make something [24:51] in 1/5if the time that is significantly [24:54] better, more polished, more reliable, [24:56] functional. This feels like a joke. [24:59] Like, anti-gravity was bad. Anti-gravity [25:02] was so bad that it caused Ben to crash [25:04] out in my video. This is comparably bad, [25:07] but it's not Google, so there's no [25:09] excuse. Oh, and look, our changes over [25:11] here finished. So let's view the diff [25:13] which actually works and renders [25:15] everything properly. I know so difficult [25:18] to do that. You can even see different [25:19] turns and what changed within them [25:21] because we snapshot all of it. You have [25:22] the separate horizontal before after [25:24] view. You can turn word rap on and off [25:26] and it just works. Yeah. Diffs.com. Huge [25:30] shout out to them for making a great [25:31] open-source solution for rendering diffs [25:33] in the browser. Also of note is how well [25:35] we handle long threads. I'm in a nightly [25:37] and I just wipe my DB so I don't have [25:39] any. But Julia showed me a thread that [25:41] literally took 10 minutes to scroll to [25:43] the top because he has been using the [25:44] same thread for weeks now for one of the [25:47] things he was working on. And the [25:48] combination of the codeex compression [25:51] and compaction being good and our render [25:54] PF being good meant that he could have [25:56] dozens of turns and millions of tokens [25:58] in one thread and scroll the whole thing [26:01] with no issues. I couldn't fathom how [26:03] long that thread was. All the hotkeys [26:06] are bound wrong. Cool. So, I'm in the [26:07] right view here in the new Cloud Code [26:09] desktop app. There's a hotkey to change [26:11] permissions. So, if you want to switch [26:12] from bypass to the DEF O, even though I [26:15] was focused on the right, the hotkey [26:17] goes left. Seems like all of the hotkeys [26:20] go to the first window except for [26:22] command W, whichever window you're in [26:25] will close unless it's the first window [26:27] and then it puts you in the new chat [26:28] view. How did they do the binding for [26:31] their hotkey layer? Julius and I spent [26:33] so much time trying to figure out the [26:34] right way to do hotkeys. And Julius [26:36] spent even more time roasting me when I [26:38] did them wrong with one of my [26:39] implementations. Little things like, "Oh [26:41] god, command 1, two, and three swap [26:43] between chat, co-work, and code." I [26:46] don't know about you guys, how often are [26:48] you switching between chat, co-work, and [26:51] code when you're working on a code [26:53] thing? Why would they hotkey that to [26:55] some of the most useful hotkeys [26:56] available? Because we have command [26:59] numbers as switching between your [27:00] threads. And notice how when I press [27:02] like I'm pressing eight right now that [27:04] they switch instantly because we [27:06] pre-render and cache properly. So [27:08] everything [ __ ] works. I'm going to [27:10] click a thread now. Takes like over a [27:12] second to load with like nothing in it. [27:15] Insane. They have routines. Create [27:18] templated routines that can be kicked [27:20] off on a schedule by API or web hook. [27:22] Cool. I can set up a cron that runs [27:24] cloud code on my machine as long as the [27:25] app's open. Great. Exactly what I [27:27] needed. You can customize with skills [27:29] and connectors. They have a whole [27:31] plug-in store. [27:34] They hid the iMessage one. No, no, they [27:37] didn't. It's in connectors. Terms of [27:38] service violation built in. Super cool. [27:41] So, they can violate Apple's toos, but I [27:43] can't violate theirs for just using [27:45] their stuff. That's great. [snorts] That [27:47] feels so shoehorned in. I thought they [27:49] had like a browser thing. Oh, preview. [27:52] Run your dev server to inspect network [27:53] requests, debug with logs, and see [27:55] changes live. That actually sounds [27:57] promising. Oh, again it has to modify [27:59] the cloud launch file because you can't [28:04] use anthropic stuff unless you want to [28:06] dedicate your project to using anthropic [28:08] stuff. It has bypass permissions on and [28:10] I still have to give it permission. Oh, [28:12] I got it to ask questions. Will this [28:14] work? Can I press one and have it do the [28:16] right thing? It doesn't even submit when [28:18] you press it. And I can't double press [28:19] enter will. Okay, enter will submit it. [28:21] Allow cla to use preview start. I have [28:23] bypassed permissions on. If this was a [28:26] vibecoded thing that two small devs at a [28:30] recent YC company were showing me, I [28:32] would probably pass on the investment [28:34] because it is so questionable. But from [28:37] a multi-billion dollar near trillion [28:40] dollar company that has supposedly been [28:42] working on this for months that have [28:43] been saying constantly that if you like [28:45] Conductor, just wait. The new Cloud Code [28:47] desktop app's going to blow you away. [28:50] Chat keeps dropping bangers for fun [28:51] intuitive behaviors like drag and drop. [28:53] Let's reorder some of these things. I'll [28:55] move T3 Chat to the top. Oh, I can't [28:57] move T3 chat. I can't change which [28:59] project is where. That's fine. I'll just [29:01] move this thread up to maybe below. [29:03] Nope. You can only drag the thread so [29:05] you can put them in pinned or split [29:07] views. Their little browser is super [29:09] [ __ ] laggy. And when you turn off [29:11] their toolbar thing here, it breaks the [29:13] page layout because now it thinks it's [29:14] wider than it actually is. That's super [29:16] cool. You can select an element and say [29:19] again, oh, I selected the same element [29:21] twice. That's great. Make this copy [29:23] better. That's a great screenshot that [29:26] it auto includes, barely making the text [29:28] readable. That copy is worse. You're [29:31] right. Let me revert it. What would you [29:32] like it to say? You didn't [ __ ] [29:34] revert it, though. I'm done. I can't do [29:37] it. I tried. I The idea of the browser [29:40] and the model having access to browser [29:42] and web logs and requests and [ __ ] to [29:44] fix things seems really cool. I've yet [29:46] to see it work well enough. This is not [29:48] it. I I can't. They I normally would [29:51] say, well, they tried, but I don't [29:53] really believe they did. I got nothing. [29:56] [sighs] If they're not going to put [29:57] effort into making the app, I'm not [29:58] going to put effort into the outro. [29:59] Goodbye. [30:02] And make sure you check out T3 Code. [30:04] It's free, open source, and doesn't suck [30:05] anywhere near this  📄 **Full text:** ![[Full Document Contents/Articles/Claude's New Cursor Killer Just Dropped.md|Claude's New Cursor Killer Just Dropped by YouTube]] ## Highlights - CLI are great, but they're not the best way to do agentic coding. That's why Anthropic finally put out an updated version of the Cloud Code desktop app, and it looks great. Wait, that's Codeex. Let me switch over to the Cloud app quick. Oh, that's T3 Code. Also looks very similar. At least you can use the Cloud models here, though, right? No worry. We'll switch over to the Cloud app now. Oh, wait. That's the new cursor glass app. Ah, okay. For real though, this is the new ([View Highlight](https://read.readwise.io/read/01krgr054qq4aqp7vcy9hhrc06)) - Tags: 👻 ai highlighted - Claude Code desktop app. It isn't a dedicated app unlike many of the others. They stuffed this in the traditional Claude app as well as co-work. So, you have the Claude chat, Claude Co-work, and Claude Code all in the same app. And it is a huge improvement from the CLI. I bet it probably even uses less RAM, though I haven't checked that yet. There's a lot to love here, and I am very thankful Anthropic's finally woken up to the fact that a slow, laggy CLI might not be the best way to do real agent coding at scale, especially when ([View Highlight](https://read.readwise.io/read/01krgr054vavfh93pr7merhmve)) - Tags: 👻 ai highlighted - guies. I have been amazed at how many people have talked [ __ ] on graphical user interfaces for their agent coding without actually trying them. It seems like developers just feel like the terminal is the end all beall for development. If you can use cloud code in the CLI, why would you ever go use the guey for it? There's a lot of reasons. Here's a simple one. Screenshot paste. And now I can see the screenshot that I just pasted. I could ask and see it here in the history and click the image and probably even save it. Oh, you can only copy it cuz their app sucks. ([View Highlight](https://read.readwise.io/read/01krgr0558b6290k5b2wfx9axh)) - Tags: 👻 ai highlighted - grin it. You just use it anyways. Despite the fact that there are dozens of better options that have more features, more functionality, more reliability, many of which are open source, too. You just don't care. And now that Anthropic is directly competing with what we have with T3 code, I will say I am a bit scared because no one's going to choose this interface. Of the current options that you have to pick from, the Cloud Code desktop app is close to the bottom, if not the bottom. But it will still become the second or ([View Highlight](https://read.readwise.io/read/01krgr0578jt8znx5gq427q50h)) - Tags: 👻 ai highlighted