This all feels like a race where the model companies try to solve doing work locally in a way that doesn't suck, before the major operating systems companies figure out AI integration into their OS that doesn't suck. It also makes me wonder why Google which has both Gemini and Android can't figure this out, and if there are lessons to draw from that.
Google is historically terrible as a product company (and has succeeded in spite of that) As their technical innovations become less of a moat (we're already there) they won't be able to win on engineering alone (they are no longer winning on engineering alone)
They are all pretty par for the course. Google used to be outstanding... but I'm not sure of a single product they have that is outstanding (def: significantly better than the competition) anymore. On the other hand I rarely use any google products these days, so maybe I'm not the one to be judging.
I'm not a fan of Google, and also not attached to Apple or Microsoft, so this isn't me trying to stan for Google, but I'd like to request that you give examples of what competing products are categorically better (and, by what metric(s) you're judging - code quality? stability? robust set of features?) -- for Gmail, Docs/Drive/etc, Google Calendar, Maps, Classroom, YouTube.
As far as I can tell, if judged by the marketplace (and breaking ties with which product I like better), Google has run away with the ball on all of those, and Gemini seems to at least be competitive.
The only major product I'd say they've sunk below acceptability on is Search, which is demonstrably dogshit now...though I suspect it's more that they have changed their definition of what Search is for, from "helping users efficiently find other websites that are useful to them" to "A convenient on-ramp to, many times per day, capture the current user intent and steer them toward something that earns Google some ad revenue."
The Pixel series outside of security (to which their own flavor of Android doesn't even take advantage of like we see with GrapheneOS) doesn't have any particular outliers that would make it any more or less enticing than another company's phone.
Their ChromeOS hardware was nice but had lackluster software and by the time it was EoL'd, never got the love of ChromeOS-present.
Google TV generally gets outpaced by onn (Walmart's brand) on cost and value proposition.
And also the fact they have shown time and time again that they just kill products over and over again.
Arguably, "exceptional" products are not ones that can vanish on a whim, like a great, great many of Google products have. Or they actually compete with other products in the same space, like a great, great many of Google products have not. Also, one would argue a good product is not one that is bought out and then deliberately destroyed to prevent its expansion into or development of a market for itself. Google is an advertising company with tremendous reach because of a handful of very aggressive and very fortunate business decisions that successfully exploded. It now uses its massive influence to exert market pressure, but the market does not always bend to its whim because sometimes it does things wrong, some of those products it pushes fail, and I can only assume some products are slaughtered because of projections on their performance regardless of their quality or utility.
It's looking like a slightly updated reskin of chromeOS with gemini features built in.
Definitely not a developer machine based on how they presented it in google IO. So if you write software, it's not looking like it'll be relevant whatsoever. I hope to be proven wrong.
as AI native developer I need VS code forks for AI to be pre-installed. Also every single command or work need to be vetted by AI by default. I am going hardcore now.
And one interesting aspect is the number of children getting these types of neutered machines as their first learning tool. I read another thread comment saying people that started with react actually feel that using straight html is more complicated. My professors say that the best textbook is the one you've read. The next generation is being indoctrinated into this way of thinking
Folks that are interested in a way of doing work locally that doesn't suck, but which integrates LLMs, may be interested in [Barnum](https://barnum-circus.github.io/). The TLDR is that it's a programming language whose frontend is a DSL in TypeScript that is well suited for managing async and parallel work, focused on control flow, from which it is easy to invoke LLMs, and which is easy for LLMs to write. I use it to autonomously ship a very large number of PRs.
I mean a couple of websites will claim 1.6gb ram on my device, what would an LLM model cause across millions of devices, when nobody even wants to use it
I absolutely want AI in my OS. I just want it to be one I can trust to serve my interests, and not the company's. I'm literally in the middle of baking one in as I type this.
The only lesson I'm taking away is that we are still very early in the AI era. AI workflows look entirely different today than they did 18 months ago and I wouldn't bet on them looking the same in 18 months from now.
The VM itself is for Claude Cowork which does all work within the VM sandbox. That doesn't help answer why they spin it up immediately and don't have a way to disable it though. Just the "why it exists" question.
If you're not going to give Claude access to anything on your machine, why are you using Desktop instead of web chat? (Real question, I don't use these much!)
At least in a corporate environment, Claude Desktop is a pretty decent compromise. Preconfigured internally deployed MCP servers and third-party connectors make many of the necessary integrations relatively easy to control.
I use Claude Code CLI myself (inside a VM, to isolate it from the host) for >90% of my needs. For the remaining fraction - email scours, cloud drive searches, other third-party connections - the desktop application is surprisingly decent. I don't even have more than half a dozen connectors enabled. In the VM I have separate, personally managed access tokens available for various third-party services. Wouldn't really try to maintain more than 5-6, otherwise it gets too confusing. [ß]
The desktop application mostly Just Works[tm] with SSO. At least when M365 doesn't suffer from their 4-times-a-day auth outage.
ß: A lot of APIs and authentication systems were designed in the stone age. You either need a 1:1 permissioned access token that can do horrendous damage, or you deal with ultra-granular, confusing and ill-designed scoping jungle where nothing makes sense. Atlassian, I'm looking at you especially. At least an MCP server, provisioned with a reasonably done service account, doesn't have all of your powers to get things wrong with.
I do use Claude Cowork and hence the VM is important, but I also leave the desktop app running all the time since I have many scheduled tasks at different times. The thing is that the VM could shutdown after being idle for some amount of time and then fire back up when you are ready to use it.
Anthropic has pretty consistently been shitty about how they roll out their software. Extreme lack of engineering rigor and thoughtfulness.
The answer is probably as simple as "no one thought not to do that."
---
I know different people work on these things so I can't do more than guess about how engineering culture cuts across teams, but given the sheer amount of carelessness and sloppiness in Anthropic's software I have to imagine they're burning investor money in training and inference because the code to do it is as bad as the rest of their software.
I didn’t get a screenshot of this, but I just found a really pointed example of Anthropics lack of craft / rush to build. If you open Claude on Windows, and click Dispatch (under cowork) to start that up, it will tell you that you need permissions windows doesn’t have. When you click the buttons for those permissions, it has broken links to macOS system preferences. I really encourage someone to try it and post the images as a reply as I am writing this from my phone.
Rule 1 with making number go up is you eliminate friction at all costs. The user's hard drive is free to you, so there's no reason to gate a feature you want them to use based on that. 98% of them will have no idea you're foisting garbage on them.
> Claude Desktop spins up a VM without no way of stopping it
I frequently make this error when I talk. My brain thinks of different ways to phrase what I want to say, but when I speak it starts with one and finishes with another. The result is almost always wrong in the way the title is, ie some variant of a double negation.
Sometimes it happens when I type, though I try to read it multiple times so often catch it.
This question is answered by the post? There is reportedly actually no way of stopping it happen. Perhaps the poster had a brain fart while typing it. Maybe they speak a different dialect of English from you.
Back in the day, personalization / customization was all the rage, as it lets the user feel the control, power and freedom. Now it's the opposite. It's about not letting user to have any control at all. I can't delete some junk apps from my phone and mac, because they are "system" apps. As a non-geek, I can't deal with complexity of the browser and account settings to stop it from what is doing. We are at the mercy of the machines.
I’ve stopped using Claude on the desktop, just because of how slow the app is to start up and interact with. It’s an absolute clunker; I’m mystified why they can’t ship something that works well given their rhetoric about ai.
They vibecoded it, and admitted as much. Once it was able to self-vibecode, that's all they did. That's why it's written in React and uses gigabytes of RAM as a chat client.
Not only did they decide to write a terminal application in React, but it's 500K lines of code. It's strange because I'm sure Claude is capable of writing a decent TUI in C. It says a lot about the engineering culture at Anthropic, at least on the software side.
Oh, a nice subthread place to vent. Their CLI is so f tragic that it is ridiculous. It keeps scrambling the terminal, scroll and basic shortcuts keep breaking, I've used so many tuis and terminal apps and many of them are a single man operation and a side project and I have never seen anything so bad.
If I didn't know from experience that directed properly claude can be powerful, knowing that they used it to create that CLI would be instant runaway based on very reasonable heuristics - if they are not able to use their product to create a decent piece of software that is not even sophisticated then it seems futile for me to try.
I just do not understand. I feel like most HN could vibe code better claude CLI in claude than the CLI (and certainly just write one) than what we have to deal with to use subscription.
Had to make a decision for a TUI I'm working on and opted for curses rather than something like textual. If I wasn't using an LLM to do some of the plumbing I'd probably have used textual to avoid the inconvenience.
There's a lot of opportunity to leverage LLMs to make codebases less bloated and less reliant on complex but human user friendly dependencies that not many people seem to be taking advantage of.
I'm with you. I have the Claude web app pinned as a PWA for quick queries, and then use the CLI for everything project-based.
I did consider experimenting with the Routines feature on the desktop app, but I'm leaning towards whipping together something with cron. I saw another poster here who has a daily PR summary routine that I think would be handy, as I have quite a few repos where I'm a sporadic contributor but would like to keep tabs.
I thought they were all in on agentic coding? They are probably just building at a surface level with only an eye towards shipping, without considering the impact of all the changes. I've seen less and less coordination between engineers as well under that model. If that's the case (Claude Code is this way). it is sort of what you get, no matter the rhetoric about "make sure to review all your changes!" It's always trade offs.
It is surprising that the Claude web app lags pretty easily when using either chromium or firefox on ubuntu linux. Chats that delay my laptop work without issues on my ipad or iphone using the app.
The web app is definitely a bit of a problem. IF there is a native app on desktop or if claude cli is much faster, i haven't tried them.
All of the LLM web interfaces have serious lag when typing after a few turns, at least on iOS safari. I’m talking seconds to start rendering input after typing or when it needs to line wrap the input.
There are lots of good answers in this thread but I think it's because they are AI companies and not UI companies. When you look at tools like AnythingLLM, OpenCode, pi, etc. you see all kinds of different interfaces, and while they might make disagreeable choices at least they do it with intentionality and direction.
They're some of the only new UIs to be made in the last decade. Almost everyone else stays in the browser (or something close like electron- claude code is actually mostly written in React, they couldn't get far from web dev). The problem is they need to interact with the local filesystem, and not many people have built apps for such a wide range of devices in a long time, and of that small talent pool I bet most are corpo coders- moving too slow and to focused on "the right way" to actually ship more than detailed Jira tickets. They also don't have time for stable releases because competition is so fierce.
But I almost always think of things from a talent-pool-first perspective. Perhaps there are actual technical issues like what Boris was referring to.
Luckily for them, every OS has (at least one) native way of building applications, and with the power of AI they could easily make 3 different desktop UIs, while reusing the same core logic.
Many people will say it’s because of the slop. I think it’s because they have no product vision. The roadmap is pretty much a random walk, which combined with the velocity of agentic coding is like digging a moat with atomic bombs.
They are moving at breakneck speed deploying on scales most of us can't even imagine. They are working in a space that's completely unexplored where getting information as quickly as possible is preferred above iterating on some feature until it's "done" while your competitor has released fifteen other features, all sucky, but one of which turns out to be a killer and makes a billion bucks overnight.
I also discovered this while
noticing my Mac was low on storage, I only clicked on cowork once and after deleting it from the folder i’m scared to
open the cowork tab coz ik it’ll just fill up the space
The weird thing is that this is probably a performance optimization for quick responses when a user asks a question.
My agent harness spins up a VM too, but it spins up on demand, cools down in 10 minutes and warms up when I focus back on the app.
The files it works on actually lives in a mount.
People take more time to type a prompt than the VM takes to spin up on a fast machine and on a slow machine, the cooldown naturally frees RAM back to the machine.
Apart from you have no idea what's going on in the VM. It's not as it has a virtual terminal. I'll play the skeptic archetype: What's not to say they're transmitting all your code back to the mother ship? Don't be naive and don't think they don't already do this.
Why not ask itself and see what it says about it. "Claude, why are you running in a virtual machine and what are you doing?".
I have two friends that are using coding agents on Windows, which was surprising to learn.
Edit: yes, with WSL2 I believe in both cases.
I would have assumed almost everyone would get a Mac/Linux computer to use coding agents because Unix is their "native" platform. It's Bash tool calls all the way down.
Does anyone know a source for reliable data on what coding agent apps devs are using? How many are using Code Claude CLI vs Claude Desktop, etc?
I mean I’m using coding agents on windows, because I’m not just going to learn a whole new operating system just to make robots write code for me.
I want tools that meet me where I’m at, not tools that demand I change up my entire UX to interact with them.
The assumption is not “what’s wrong with Windows that it doesn’t work with <technology>,” more “what’s wrong with <technology> that it doesn’t work with Windows”
Why wouldn’t you want your thing to be cross platform
and on my Mac any time I accidentally click Cowork which I don't use whatsoever, it re-makes the same VM, without asking me. It's one of the dumbest things ever. You're about to hijack nearly 20GB of my storage (which gets eaten up as it is) and you don't think to ask me if I even want the VM before you shove one into my system?
reply