Zed is incredibly snappy to open projects and search files but very basic functions like auto-complete (at least in Python and markdown) are still terrible, nowhere close to IntelliJ/Pycharm. Or maybe there is a very specific settings.json incantation that I am missing.
I’d like to know any IDE that rivals Pycharm for Python autocomplete quality. (I am genuinely curious, because I would prefer not to use Pycharm.) My understanding is that Jetbrains rolled their own LSP and no one else has yet matched it in quality.
Though I don’t understand what you mean by markdown autocomplete.
That's why I still use two editors: IntelliJ for when doing "serious" work and Sublime when I need to edit random files or a huge JSON. I don't need anything in between.
GitButler created huge problems for me twice - it automatically added files to commits, including one that contained secrets. Okay, I know I had to add them to .gitignore, but why didn't it prompt me to add the files? There were even logs and cache files, among other things.
For me it's.. okay it's my daily driver already, but I really really want extensions to be able to create their own UI elements in the buffer, like VSCode does. Basically GPUI for extensions.
This would unblock people to write their own Jupyter integration for example, or whatever else they want. There's load of cool stuff like Argus https://github.com/cognitive-engineering-lab/argus that rely on creating buffers with custom UI, and Flowistry https://github.com/willcrichton/flowistry that rely on graying out some code, and I want this stuff on Zed too
A git worktree directory contains a .git that just references the original directory’s .git, and Zed doesn’t support this configuration. So, there just isn’t any representation of change tracking when working in a worktree directory.
Gotcha. Is there a request in for this? The team seems incredibly productive (I'm sometimes offered multiple updates per day), and my completely uninformed and naive take is that this probably wouldn't be too big of a lift, relative to the stuff I'm seeing them ship regularly.
Two things. First, some economists study stated versus revealed preferences. [1] The idea is to figure out what people do rather than what they say they will do.
Second, in the case of people making feature requests, it could be a net-societal-gain [2] if feature requesters made some kind of binding commitment. (See also the hold-up problem [3].) Perhaps a potential customer would commit to "if/when feature X gets added, I will commit to using the product for 2 hours." or "... I will spend $10 on the associated cloud services." (The question of what happens if the customer reneges also has to be agreed upon up front.)
Okay, so what kind of solution are you looking for here? VS Code uses a closed-source LSP server for its C# extension. Rider is it's own custom stuff, of course.
So...where does that leave the Zed team? If existing LSPs aren't good enough, that's not a Zed problem: they're building an editor, not LSPs for your favorite language.
I tried it out this morning and it felt really rough, unfortunately.
It was super slow (thought I think that applies to CLI Codex too), it wasn't outputting any text explaining what it was trying to achieve, and it started off down a path that made no sense. Claude Code in Zed has some rough edges but it's at least usable.
In terms of GUI agents, Cursor is still a lot nicer experience, IMO. Though I do still prefer just using Claude Code cli, personally.
I have been using a handrolled integration in zed for codex and that is really slow compared to claude code. I think its just the nature of the codex beast.
Can someone help me understand the pricing of zed? $10 per month- $5 credits for AI credits. This credits can be used for claude code / codex inside zed or should I manage different api keys for codex/claude code?
There are 2 modes of operation - an editor AI mode and a dedicated agent mode. For the agent mode like Claude Code or Codex, you don’t have to pay Zed, only the CLI tool providers. The zed subscription is for those who don’t want to deal with AI vendors, cli tools etc., and just use it in the editor
The speed at which this team ships is genuinely insane. I'm already expecting the next blog post to say "We've added fully native support for the M3 Max, Windows, and a built-in neural net for generating feature flag names. Oh, and here's the Zed 1.0 release candidate.
As much as I like this initiative (I'm a Zed user), what is available in the UI as Claude Code is not really Claude Code. It's extremely limited in terms of context management, for example. ACP is reducing these tools to some shared functionality, but that makes them less useful.
I'd wish a better model or system would go live for the inline suggestions. The Zed ones are so trash compared to Cursor's it is just laughable.
An example is that when I have a module like Namespace::SuperAbcModule in a file at namespace/super_abc_module.rb and I rename the file to namespace/super_module.rb, Cursor will immediately suggest to change the module name to `Namespace::SuperModule`, Zed won't.
Also Cursor will suggest updates to lines throughout a file whereas Zed sometimes doesn't even look ahead 1-2 lines.
Having Claude Code and Codex built into the sidebar is hardly better than having them running in a terminal. I wish they'd invested all this time and effort improving the inline suggestions.
Right click on the file in the project tree -> rename will rename references. Or in the code right click -> rename symbol. Not sure why you need to bring AI into it.
Why is that something AI has to do? PhpStorm does this, without Ai, since forever ago. And updates references everywhere, even inside strings or doc comments.
Firstly, I don't care what tech does this, as long as I can get good suggestions.
Secondly, it was just one example that came to me from comparing this the other day. You could compile a long list of examples where Cursor gives better completions than Zed does.
It's better to learn the tools that you are using. Renaming is part of LSP servers and Zed has no control over them. They are unique for each language. Using AI for this is a waste of resources, especially when renaming affects tens of files. It's not reliable, slow and not determenstic.
They were just giving that as an example that Zed's inline suggestions aren't very good for basic tasks. There are hundreds of othersmall tasks like this that can't be handled by the language server.
Yes, elsewhere in this thread someone is complaining about lousy C# language server performance relative to IDEs. These swiss-army-knife programmer's editors will always be at a semantic language tooling disadvantage relative to IDEs.
I know that days of yak-shaving with LSP and emacs only gets me to a janky imitation of Visual Studio/XCode semantic search on my C++ work codebase. "Fuck it, let an LLM auto-complete based on vibes" has some appeal when you just get sick of trying to arm-wrestle clangd into ... whatever XCode or Visual Studio are doing to have functional semantic search across the project.
Although I have to say LLMs were a disaster at vibe-auto-completing in VSCode. So I mostly stick with semantic search in the IDE and editing in emacs like I always have.
I otherwise like Zed way more than the vscode-derivatives but yeah, the edit predictions are just not even close. And it's laggier feeling despite the lower quality.
While your example is not AI related (should be handled by the LSP integration with "Rename Symbol") I agree that Zed's Next Edit Prediction model is *extremely* subpar. Imho they should either scrap it and just work on having a good integration story with third party models for the next edit (and maybe propose by default a partner model I don't know) or invest a lot more efforts into it.
But currently I sadly have to say the model's "help" is often a net negative.
It's human nature to start trusting AI suggestions just because they look good enough without actually checking them. That's going to lead to massive issues down the line the more it goes on.
Snippets are more useful.
If you're doing something repetitive, a vetted snippet does wonders. Learning how to make your own snippets with appropriate tab stops is a seriously underrated skill.
High competence in regex search-and-replace, multi-cursor, and snippets is an amazing combination.
Improved inline suggestions are under way if you search the PRs on GitHub for zeta2. You can also bring your own edit prediction provider by configuring copilot or supermaven. Codestral edit predictions were merged last week
Keeping data out of Zed’s servers and letting users handle OpenAI billing is a smart trust move—too many tools bury those details, which erodes confidence fast.
Sorry, I did indeed use AI, the main reason being that my subject is not in English, so I had AI help adjust it appropriately. (This is a direct translation through Google Translate.)
Zed has a generic interface to agents (ACP). This is a bridge between the ACP api and the codex api so that it integrates cleanly inside Zed in mostly the same way the other non-codex supported agents do
I'd like to know this. I really like Zed, but their paid AI thing is terrible at inline suggestions, at least for my use cases. At least Copilot's C was coherent, even if it was wrong all the time.
reply