I find the benefits of lightweight backpacking type optimization overblown for my travel style. Maybe it’s important if you’re speed running 20 destinations in a single trip or are incredibly unlucky with checked bags getting lost, but especially with kids now I just don’t really see much benefit.

- Composer 2.5 is cheap, fast, and very effective; I don't even use other models that much anymore, as it's usually marginally better for way more cost, though sometimes I do for specific things like making better translations for our app than a coding-specific model could normally output

- It makes setting up and maintaining Cloud Agents super easy, the agent can basically set up itself and if anything changes that makes it not update properly, then it tells you and can fix itself easily

- Easy to move agents from remotely in the cloud to local and vice versa

- Easy to work with agents via either comments on GitHub, or via the web app on my phone (though it is relatively constrained relative to the actual desktop UI, which is a bummer)

- Code reviews with Bugbot is surprisingly good now vs when it first came out, as is the Security Agent, while being an order of magnitude cheaper than stuff like Claude reviews

- Automations are easy to configure and manage - crons, in response to repo events, etc. For example I don't use Renovate or Dependabot much anymore since LLMs can update deps and investigate subtle breaking changes much better than a dumb version bump script can

- Limits are obvious rather than Anthropic's mysterious quota amount that they don't explain at all

- Queueing up messages rather than the agent taking in new messages mid-work and then trying to mesh them together somehow - I find queueing much more predictable and easier to work with

- Plan mode is good too, but not particularly different than any other agent

- Easy to jump into the Editor view and actually go in and manually code things or interactively code with the LLM when you need to, since sometimes LLMs just suck at doing certain things autonomously. I'm not in the "stop coding bro" camp, I still like to take the wheel fairly often.

- I tried to use the Cursor Agent via ACP, it worked but it seemed markedly stupider than when I would use Cursor directly (saying that i18n strings were being used when they weren't, editing code differently than what I asked for, also when it is running terminal commands it seems to just say "Run Command: Terminal" and has no information on what's going on). Maybe I just need to not use Cursor Agent, but my company pays for it already so that's what I tried.

- Providing context is also cludgier - In cursor I often highlight specific chunks of code and add it to the AI context via Cmd+L, but I couldn't figure out any specific way of doing that with a keyboard in Zed besides clicking a tiny + button to add "Selection" to context, which got old fast.

- Maybe I just need to get used to it but reviewing code with the git integration is just hard for me to follow; one giant editor with every change in it is just harder for me to grok than showing each file one by one; so it was tiring to review the big changes produced by LLMs. Also, when you stage changes the file just stays where it is with a barely noticeable check in the checkbox in the sidebar - I prefer the behavior of Cursor where it actually moves the staged files to a separate section, but some kind of more obvious visual indication besides that perhaps would help. I did like the tree view, though!

- The tsgo and oxlint LSP servers kept crashing, which was frustrating. GraphQL LSP server also couldn't understand graphql.config.mjs, which is strange as that's supported out of box by graphql-config and works fine in Cursor/VSCode.

- I tried using a few of the different Edit Suggestion LLM options, but unfortunately Cursor is just way too good compared to any of them (slow, and just not very helpful in comparison).

- Just in general figuring out how to configure them is confusing, there's like 3-4 different places to configure agents and LLMs for different purposes, I found it very fragmented and confusing and the docs didn't make it particularly easy to set things up.

That all said, the performance was muuuuch better than Cursor. But the UX issues and general bugginess of ACP and these LSP plugins were impeding my workflow too much for me to tolerate it so back to Cursor for me. If anyone has tips on how I might make some of these better that would be cool, but if it's just inherent limitations then maybe I'll try again in 6 months or something.

If the incentives are right American companies can make good things, but usually they are not so because of poor policy.

Unfortunately, people from western countries have very negative views toward the privatization of mass transit despite the wild success that Japan has experienced. The model makes so much sense: if trains are just a way to get people to the real estate that you developed, then you’re going to make sure that the trains AND the destinations are really nice, which also turns out to be very lucrative (at least in densely populated areas) as a cherry on top.

And even worse, like this commenter above alludes to, it is trendy in the West to believe that real estate developers are evil, and that corporations that make money are sucking the life out of society. This kind of degrowth populism pretty much guarantees that the successful Japanese model is out of reach for most countries, because it is exactly the pursuit of profit that makes Japan’s system so nice - not some edicts from a benevolent and extremely capable government.

[1] https://news.ycombinator.com/item?id=47762060

That said this reply doesn’t actually address much of what the article talks about, most interestingly how rail companies are private and are also real estate developers. That thought process ought to make sense to Texans or something.