Zed was the first editor that tempted me into using AI features. It felt solid in general and AI feels mostly like autocomplete in other editors (in terms of how much it's in your face). There's definitely a place for AI models and agents in code editors, and Zed makes me feel like it's not built around them, which is great! Zed feel like "Come to us, we are making a good fast editor that also has AI." while competition feel like "Come to us, we want AI that has an editor".

I'm genuinely happy it works for you. I just don't want AI in my text editor, even if you're happy with it.

I suspect you’ll be an extreme minority.

I won’t go near a code editor anymore without AI integrated deeply.

I suspect I'm not the only one experiencing "AI fatigue" as every single piece of software grows more and more useless "AI" features I don't want and which get in the way of doing what I actually want.

Yep, I was on mailing list etc. Unsubscribed when they did the AI announcement. Its not for me.

Nah, its the quiet majority, and AI posts on places like HN quickly create their own echo chambers with folks patting themselves and each other on their backs.

There is AI as a useful tool, maybe, which is at most few % of current hype. Most folks seem to end up babysitting it a lot to get something useful out of it. And then there is everything else which is mostly hype or narrow use cases. To proper typical senior managing a team I don't see much added value. It can help juniors churn out large chunks of the code but I haven't worked in 20 years in a place that values quantity of code and quick deliveries over quality.

Also very much depends on the business and specific company. In my banking mega corp, no AI is even allowed to be used even as I write it now, all popular sites are blocked and there are strict policies against. Couldn't care less, coding is such a small part of my work I don't want to lose this creative outlet by delegating it to something I need to triple check for bugs afterwards. Also with any new stuff I learn way more by implementing it myself rather than looking at pre-made code.

> Also very much depends on the business and specific company. In my banking mega corp, no AI is even allowed to be used

This is a huge thing tbh. I don't like these AI things in general so I wouldn't use them anyway, but I just can't imagine going to my clients and asking them, "Hey is it okay with you if I routinely upload all your code to these random American venture-backed start-ups?". And I really can't imagine just doing that behind their back. I couldn't really imagine doing that with an employer's code either.

Ideally I don't even use software where accidentally toggling the wrong checkbox in some settings screen results in automatically uploading client code to these American start-ups either. Now I won't pretend that my stance against AI is purely out of some principled cybersecurity concern, but it's definitely a factor.

You can use local LLMs.

Is that what Zed's integration does by default?

In any case, what I see most people doing is integrating it with Claude/Copilot/etc. The security concerns specifically obviously don't apply when running it locally.

By default Zed has a user choose what type of configuration they want. That could be something like LM Studio, Ollama, or your own API keys to a provider you're already paying for.

AI also isn't shoved in your face when using Zed, there's one small button on the button right.

> there's one small button on the [bottom] right.

Wait there's an always-present AI button in the normal text editing UI? That's way more prominent than I expected, I assumed it was just an option in a settings screen somewhere. I definitely don't want an AI button that's always on screen.

I just downloaded Zed to see this for myself and found not only one but two AI buttons in the lower right, one for integration with chat bots and one for their "prediction" AI. Both try to get you to log in to online services (even though, yes, a local chat bot is an option for one of them).

It seems like you can remove the chat bot ("agent") button through their config file, but I found no option to remove the "predictions" button.

Man this editor is pushing "AI" way harder than I imagined. As I said I genuinely assumed that it was just like iTerm2's chat bot integration where you could enable it in a settings screen.

AI features are essentially non-negotiable for a lot of people these days, and a lot of companies too. It’s hard to deny that LLMs coding assistants are at a minimum pretty promising. More promising than many other LLM uses. Zed wants to be a professional IDE, not just a random text editor, and LLM services are (whether you like it or not) an important tool in that IDE toolbox.

It’s fair to have huge concerns over AI (I certainly do), but I don’t think it’s fair to expect IDEs to skip some of the only technology that developers (and companies, importantly) are willing to pay for in an IDE.

I strongly agree about the fatigue. I have extreme levels of concern, especially when the supposed top people are shooting their mouths off saying this stuff is doing material science, biochem, a million other things.

But all that annoying madness is distracting from how amazingly awesomely useful this stuff is. It's wild how many little quick projects I can kick out in a couple hours! Ideas just come out of my head with so much less fuss; when I don't like it I ask for something different.

My point is less to convince though about AI. I appreciate your starting sentiment here, but I really don't get the follow up?

> genuinely happy it works for you. I just don't want AI in my text editor, even if you're happy with it.

I don't see why it would bother you at all? There's a tiny little button in the status bar and a few scattered menu items that feel, to me, very easy to ignore.

It feels like someone being mad that their spreadsheet has I dunno, logarithms in it, but the person hates logarithms? It feels weird to opt in to caring against. I have a generalll abnner of thought which is "your anti-feature is not a feature", and this feels like one of those situations: i don't see why someone would cling to an editor not having a feature they don't use?

And I suspect you are an extreme minority. Among senior and staff engineers I barely see them leaving behind emacs, vim and jetbrains… none of them with ai plug-ins.

Let me clarify. Things are only going in one direction.

So if you’re not in the minority now, you will be in time.

This is not a “tech bro” thing (as someone else said). There is real substance to it.

Nope, not a minority at all. Quite the opposite outside the tech bro side of things.

I suspect GP will not be an extreme minority.

Then don't use it? It's completely optional and works perfectly fine without AI

Thanks for the tip, but I was already not using it

It's an unexpected position to take, though. You said you had been interested in Zed until they integrated AI. The response was "the AI is completely optional", which I'd expect would make you more likely to use Zed, since it removes your objection. But it doesn't change your position at all, which makes me suspect it's not that you're worried the AI would interfere with your workflow, but that it's there at all. So, is your position that the very fact that Zed allowed AI to touch it has infected it in some way?

I'm not interested in using text editors with chat bot integration, hope this clarifies things

Not at all!

So you're just here to be an AI curmudgeon with no valuable input to the debugger conversation topic whatsoever, with no experience with Zed and no intent to ever even try it? Thanks for all the relevant and useful hot takes I guess.

The way I see it, I shared thoughts on Zed, as someone who was once really interested in Zed and once tried to switch to it, in a discussion about Zed. It's not about the Zed debugger, but it's not like this thread is lacking in discussion of the debugger.

Am I an AI curmudgeon? I wouldn't necessarily use that word, but it's not entirely inaccurate.

I went to check out neovim and noticed it's currently sponsored by two AI products! Of course, that's one level removed from actually integrating AI in your product but, still—it's getting harder and harder to avoid altogether.

Oh wow, I hadn't noticed that.

I guess it's always possible to return to Vim if Neovim starts showing signs of being steered by its sponsors.

At this rate you're going to be cooking over a campfire and living in a cave in a few years.

Nah, the AI bubble will have popped in a few years and projects will stop sprouting AI features left and right.

Did the dot com crash cause software to stop integrating internet features?

No, we still have software with Internet features. But Internet features went from a solution looking for a problem, to just another tool we can reach for when it's actually useful.

A whole lot of "AI" features today are in that "solution looking for a problem" category. There's a lot of "AI" in places where it really makes no sense at all. Companies and projects are afraid of missing out on what they think could be the Next Big Thing, instead of just trying to make the best software they can.

When the AI bubble bursts, it could end up like Internet features: software gets them when it genuinely makes sense, but they won't be crammed into software which has no need for it. Or it could end up like cryptocurrency: it pretty much just disappears as people realize that they don't really have any use other than to speculate on its value and to buy drugs.

Personally, my bet is that they'll end up more like cryptocurrencies. After all, "AI" doesn't just have to be a useful feature to be worth it. It has a real cost. Companies like Microsoft and Apple and Google, as well as the venture capitalists and investment funds behind the likes of Anthropic, are currently sinking VAST amounts of capital into giving "AI" away for free or heavily subsidized. At some point, it'll need to become profitable, and I don't think many people will find that the value outweighs the real, non-subsidized costs.

But we'll see.

The Network was a good idea fifty years ago, and it was still a good idea by the time the Internet was definitely the next network which was a few years before the DotCom bubble, both actual large language models and the "What if all human conversations in a box?" thought experiment (this definitely passes the Turing test, but, it's clearly not a person is it) are also good ideas, distinct from present "AI in everything" LLM craziness.

Just because tulips are nice doesn't make Tulip Madness a good idea. At the height of the DotCom boom consumers were buying stock at IPO prices for companies which made no sense whatsoever, because they said "Internet" or, (hence the naming) ".com"

For example Be Inc. was a vehicle for a failed Apple exec to "prove" he was the right person to run Apple, not Steve Jobs. After their runway ran out and institutions wanted nothing further to do with it, Be Inc. went IPO. They do this by saying they were an "Internet appliance" company and taking an OS with terrible networking but pretending it's good. In normal times this would attract laughter - they're offering a worse product, most likely it just tanks or never comes to market, and in the extreme case that Apple wants the CEO they're going to cut a deal with the man, not save the dead weight company. His most senior people might get parachutes but Apple has no reason to pay ordinary stock owners a penny. Sure enough those who bought at IPO rescued the institutions but were wiped out.

My guess is that neovim wouldn't do a core integration of ai that ships with the editor. At most, maybe if there are ui interface gaps that would help make a better experience, they might expose more apis there that could be useful outside ai as well.

Plugins are relatively easy to write in nvim so I'd expect all ai stuff to come from there and be opt in.

Thankfully for me, I guess, neovim broke my config about a year ago by changing the default color scheme in a way that I could not fix easily with confug. So I forked and built my own and will likely never update it.

I didn't tried Zed in a while, are really intrusive the AI features? Can't it be just disabled with some configuration?

they are not intrusive but their entire focus changed on that instead of other features. Entire Git view feels abandoned in half done state yet they spent entire month working on AI chats, AI agents, their own AI edit (that's priced 20 per month yet they boast how light and performant it is -- why isn't it free local model then and why its priced worse than copilot?)

They're moving from "making awesome code editor" into yet another "buy our ai" product

Wouldn’t the addition of the debugger disprove this slant? Huge feature that has nothing to do with AI.

debugger is very far from being feature-complete, I would call it MVP at the moment; lets see if they will iterate on that or will quickly go back to new shiny thing...

AFAIK they added conflict resolution just recently, so it's not like non-AI features such as Git get no attention at all. And of course the debugger now.

I'm much prefer the VS Code style when you have 2 clear sections: staged and not staged. Zed's chose the IntelliJ style which is just a bunch of checkboxes, I can see it being easier to understand for the novice, but not very intuitive from a git point of view.

Never used Git in Jetbrain IDEs or Zed (yet), but recently VS Code improved a lot the Git by making easier to stage changes in a file in a granular manner (I tried briefly edamagit extension a while ago and now use a lot gitu in the terminal, in combination with VS Code version control).

The fact that they have released both features within 3 months of each other is mind boggling. Their development velocity is insane. These are not trivial features.

Give them some time to polish, jeez.

Copying a buffer into a network call, reading a response, writing the buffer. Not trivial per se, but table stakes for an experienced developer.

Rust probably slows them down here, but working correctly early is preferable imho.

Over simplify much?

Which computer science frontiers are being pushed here at the user interface level, do you think?

(This is referring to their recent integration work. The acceleration layer was usable a year or more ago.)

I suspect their previous “collab editing” marketing angle was probably not a big enough draw. AI features seem to be desired by more people, or at least the hype cycle currently says so.

We have multiple teams at Zed each with their own focus. In fact, you can see all the projects we're working on if you view our channel notes.

We're going to be adding more features to the debugger for a while

I think you're referring to autocomplete? It's much better than, say, 2 years ago when it was indeed annoying as hell. Having said I always turn it off and use agentic coding, which is not intrusive, only activate if you ask for it. This applies to all coding tools these days, autocomplete is no longer their focus.

I don't know, I just uninstalled Zed when I read those features got added

Me too. I don't want AI, and if it's there, I want to be able to completely remove it. Zed is forcing it, so I'm staying on VSCodium.

What's forcing AI in Zed though?

agent.enabled = false and it's gone, no?

Probably also requires an api key, no? So would be difficult to force.

I have a firewall, OpenSnitch, so don’t have to worry much about programs trying to connect. But definitely would prefer they don’t unless directed.

How does zed handle it?

There's a pair of config settings to turn off all of the AI features:

features.edit_prediction_provider=none agent.enabled=false

There's additional config to set if inline assistance is automatic, only when the user presses a key, fully disabled, etc.

I’d prefer an option at first start perhaps, to turn it on.

That might be how it works, I haven't been through the new user flow in a while.

I find it very easy to avoid the AI feature in every day Zed usage. Sometimes they do come in handy though. But nog often.

I don't want an editor where I have to avoid the AI features, I want an editor without AI features

Just like I want a terminal without AI features, which is why I'm no longer using iTerm2

I didn't even notice iTerm has AI features? Where?

It seems like they've since removed it from the core app and put it in a separate plug-in: https://gitlab.com/gnachman/iterm2/-/issues/11470#note_19176...

Anyway, I'm happy with Ghostty since I switched away from iTerm2 and haven't paid attention to iTerm2 development much.

The developer of Ghostty is using AI for development. Better quickly delete it!

As long as he doesn't get tempted to add AI features into Ghostty I'm not too concerned

So you're interested only in tools that do nothing new and offer no added value on top of a normal code editor? Plenty of choices already, why even interested in any new thing, ever.

To be honest, I'm happy with running neovim in a terminal emulator so I'm not that far away from just wanting tools that "do nothing new". But there have been monumental changes in my time as a programmer that I've liked: language servers have been huge and are an essential part of my programming experience now, as one example.

But I think you're under-estimating the value of just being a really good version of something that already exists. There was a potential path where Zed would've been more or less something like VSCode, but actually open source, with less jank, with a smaller footprint and better performance, with excellent out-of-the-box support for a selection of programming languages without needing extensions. That, to me, would have been revolutionary, not because it would've had a list of features that's larger than other offerings, but because it would've done everything better than existing software.

Besides, it's not like "AI" is some new thing that nobody has done before. Every single new text editor that's coming out seems to be selling itself as "AI-powered". Zed is now a follower in that crowded space.

Separate tools that do one thing well, works better sometimes.

E.g. editors often integrate a terminal and file manager. I already have them open constantly, so have never used the integrated versions.

I used to use separate of each, but switched to the ones integrated into Zed. being able to click on a file in one and have it open one frame later in the others is very productive for me. e.g. click a link in my test runner's output to a line in a file and immediately jump to that file in the file tree and that line in a buffer.

Hmm, open tab performance hasn't been a concern of mine... since the turn of the century? (cough) Not using electron of course.

I suspect a preference here has more to do with how many monitors one has. I have multiple so prefer multiple windows. If away from my desk, having a fullscreen window split makes sense.

Yeah, in my office I use a single ultrawide. And I'm often programming away from a desk on a laptop.

Back in the day I had a multi-monitor setup and that was super nice with a tiling VM and a terminal/*nix based development environment.

The AI features in Zed are very easy to turn off / ignore. I agree that the AI features are probably taking development time away from other features that might be more useful.

I just disabled everything AI related. It's a nice editor. But I still have to jump into VSCode to resolve merge conflicts.

Recently the Zed team added some tools to help with merge conflicts: https://github.com/zed-industries/zed/issues/4964#issuecomme...

Yeah, but still it's not as comfortable as with VSCode or for example Meld.

You can disable AI.

Furthermore, Zed Agents are currently my favorite way of using LLM's during programming

It's trivial to disable it. I wouldn't let that hold you back from using an amazing editor!

I just tried it out again. It seems like you can disable the chat bot integration by adding this to the config:

    "agent": {
        "enabled": false
    }

However from the documentation[1] I can't see a way to disable the "AI" predictions button (which asks you to sign in to their online "AI" service with your GitHub account). Am I missing something?

[1] https://zed.dev/docs/configuring-zed#edit-predictions

Setting this to none should remove that button. "features": { "edit_prediction_provider": "none" }

Really? That's not mentioned in the documentation, it seems like the whole "features" object is missing. Thanks anyway.

Relevant GH issue: https://github.com/zed-industries/zed/issues/6756

how does editor neovim connect to debugger in the post?