Zed: The Open Source Agentic IDE - Use Claude Code, Codex & Gemini CLI in one place

--- type: transcript date: 2025-11-25 youtube_id: QU4hED-RZ5U --- # Transcript: Zed: The Open Source Agentic IDE - Use Claude Code, Codex & Gemini CLI in one place In this video, I'm going to be showing you Zed, which is a powerful open source editor. They just recently released something called agent client protocol. It's a new open source standard and what it allows you to do is to integrate different agent harnesses directly from the major labs. Think things like cloud code, codeex or Gemini CLI. One of the nice things with this and where it is a little bit of a gamecher is you can with your pre-existing subscription to those different services directly integrate them within zed. If you want to have both Claude code or say if you want to have Claude code one month and then Codeex the next month, you can keep the same editor but swap between these different model harnesses. As new models come out, they're constantly leaprogging one another. Like for instance, if you want to use Gemini 3, you'll be able to swap to the Gemini CLI. Or if you're like me and you generally prefer something like Claude Code, you can sort of use that as your daily driver and maybe just experiment or dabble in some of these other harnesses as well. And one of the great things with Zed is you can take the same powerful agent, but bring it in a professional IDE. It doesn't just have to live within a CLI. It can be sort of within the experience that I'd argue most people are probably most familiar with. And not to mention one that is actually performant and one that's not actually based on Electron and doesn't have some of the issues that come with that. One of the benefits of Zed is it's built from Rust from the ground up. Just yesterday when I was leveraging Cursor, I got this message twice. Now, mind you, I had a lot of different windows open. I probably also have a decent number of extensions installed that have just sort of added up over time. And that is something that I have seen as well as others have seen with some of these electron-based editors. So, in this video, what I'm going to do is I'm going to walk you through a demonstration. I'm not trying to build a spaceship here. We're not going to be doing anything super complicated or low-level, but I'll just walk you through a little bit of a build, show you some of the Aenta capabilities, and specifically just focused on some of the features of the editor and highlight some of its aspects as well as how you can get started. It is super straightforward. Let's dive in. First things first, you can go ahead and download. It's available on Mac, Linux, or Windows. You can also go ahead and check out their open source repository. I do encourage everyone to also star the repo as well. Definitely support open source where you can. I just want to touch briefly on agent client protocols because this is one of the key aspects of one of the things that they built here. The agent client protocol as they describe is it standardizes the communication between code editors and idees and the respective coding agent that you want to use. And you might be wondering okay why actually leverage this technology. It is going to become very clear as I show you within the demonstration. But one of the key aspects of this is every new agent editor combination requires custom work. One of the things that you don't have to worry about is integration overhead, limited compatibility, as well as developer lockin. If you want to try, say, if Gemini 3 comes out this week and you want to use it, but you're sort of bound to having everything within something like Cloud Code, this allows you to have a little bit more flexibility. So, you can have the same environment of what's within Zed, but you can easily just swap to a different agent. And as well, if you want to use just the API keys of different providers outside of some of these agent harnesses, you can also leverage that. But that is a little bit of an aside from the overall agent client protocol. And the cool thing with agent client protocol is it's actually being adopted across even what you might consider competitors to zed. Things like Auggie from augment code as well as Jet Brains and their family of idees. That's one of the things that I definitely appreciate with Zed is there definitely is that culture and vibe of open source where they're putting things out there that's actually going to benefit some of their competitors. But one of the benefits of that is at the end of the day the customers ultimately win. It is a really nice thing that there is support for this. You can go ahead and install it on whatever machine that you have. You can go to the download page and get started very easily here. Once you have it installed, this is what it looks like. It has a lot of the familiar commands if you're moving from something like VS Code or cursor. You're going to be able to open and close your different sidebars as well as your terminal with the similar key bindings. Additionally, you're going to be able to have the agent window on the right hand side like we typically see with these integrated AIE type of environments. I'm going to open up a new directory. And within here, what I'm going to do is right off the bat, I'm just going to create a new Nex.js project. Just to demonstrate this, I'm going to show you a web app. What I'm going to do is I'm just going to go ahead and kick off the defaults for installing Nex.js. And then once I have that, I can command shift question mark. And that will open up our terminal here on the lefth hand side. So what you'll notice within here is we do have that cloud logo in the top leftand corner. But what's really great with this is let's say for instance I want to swap to the codec. It's as easy as simply opening that up in a new thread and then what it will do is it will spin it up. You can almost imagine as if running that command to run codecs or claw code. Effectively, when you start a new conversation, you can think of it as running that command to actually initialize what you would have otherwise done within the terminal. And then from there, you can begin to interact with it. So the first thing that I'm going to do is on the homepage, I wanted to read developers digest. And specifically, I want a neo brutalist style that has primarily black and white as the main colors. Once I have that, I'm going to go ahead and send this in. Now, one of the cool aspects of how they've incorporated this is you can also follow cloud codes. For instance, if you have used it from the CLI, one common complaint that a lot of people have is you don't have the visibility into what it's actually doing. And that's also something that I do notice within Cursor's new composer. It's actually going through things often very quickly where it's streaming out all of the different tokens, but it is a little bit overwhelming. And the nice thing with actually being able to track changes like this is it's not trying to show you look how fast this is or anything like that. that actually gives you a moment in between all of the edits when it's going through and actually generating things. So, here is what it has generated for us. A simple little web app, but let's just circle back to the actual conversation thread. The key with this is all of the different commands that it's running where so one of the key aspects of this when it's looking through the different directories trying to understand our project and then ultimately start to make edits and make decisions in terms of what it actually needs to do. This is just like you would be running the CLI. you have that exact performance but with just that augmented experience of it being within this nice UI as well as this performant editor within Z. It does make a pretty compelling option to be honest. Like I know a lot of people newly like the CLI. It's sort of a new sort of popular trend especially since cloud code came out. But honestly I think in terms of actual professional development this just makes it a little bit more practical to use and there are a number of helpful benefits of having it within this view. Some of the things that I appreciate with Zed is having used cloud code for some time is there are a lot of really nice nuances and UX decisions in terms of how they actually brought this out. For instance, cla code if you've leveraged it and you ask a task that requires say a complex plan. What it will do is it will go through that plan iteratively and go down that list. And how it works within here is we have this nice area where we have the plan within here and we'll have all the different steps if there was a multi-step plan within here. And additionally, it will show you all the different files that were edited. And then you similar to something like cursor or one of these co-pilot type of tools you can go ahead and accept all of those changes just like that. Now the other thing that I do want to mention with this is you can atmention different files if you specifically want to use different pieces of context. Let's say I want to grab that page context. I can leverage that. You can also reference symbols. Where this is helpful is let's say I want to reference something like the layout or the different props somewhere or the different routes that might be set up. You can also reference threads, you can set rules or you can also fetch from the internet as well. Now, additionally within here, you do have the option to add in your own custom MCP servers. For instance, if I want to say I want to research the latest AI news and reference those as the articles. One of the things that you'll notice with this query is it's asking us if we can use the fire crawl search MCP tool. I'll go ahead and click always allow. Another aspect of this that is nice within the UI is you can switch between the different levels, how much permission you want to give to your agent. Say for instance, if you want to bypass all of the different permissions, let's say you have a large refactor and you just want it to just effectively accept all of the different decisions that it makes, you can go ahead and bypass all of those different permissions. So it won't ask you each time when you want to run different MCP servers or what have you. Next up, I want to show you some of the MCP capabilities. Now, what I wanted to do is I want to within the different blog posts, I want to have a blog post on Gemini 3, Sonnet 4.5, as well as GPT 5.1. All of those just came out. Make sure to research all of those different topics. Just make sure that it does leverage my MCP tool is I'll say leverage the Fire Crawl MCP for search as well as scrape. As soon as I send that in, what we'll see is since I do already have the MCP for Firecrawl all configured here, it will ask for permissions when we go ahead and use that tool. Now the other aside that I do want to point out within this is depending on the different mode that you want to have similar within cloud code you can go between the different modes where it always asks you or the plan mode. Now one of the nice things with this is actually being able to set the bypass mode where say for things like refactors or if it's a low stakes application or if you just want to trust the agentic process you can enable that bypass permissions and it won't ask you before it invokes different MCP tools or writes to different files or what have you. It's a little bit more risky for the bypass permissions mode, but for some tasks, especially if you're very clear in your instructions and you're a little bit mindful of some of the considerations of making sure that it doesn't actually run different commands that you don't want it to run, that bypass permissions mode definitely can be quite useful. Now, Claude's asking me for some clarification here. And what's interesting with this is it didn't just go on and proceed to go through the task through and through. It actually paused and it said, "Okay, I have to be transparent with you. Here's what I found. did find information about GPD 5.1, but it found Gemini 3.5, but it was not verified as an official release. Within here, it has a recommendation. I can write a comprehensive blog about GPD5 as well as well doumented and official sources. However, I cannot in a good conscious write about Gemini 3.5 as a release product since I found no credible evidence it exists. That honestly is actually a really nice answer from a model because oftent times models at least until recently they would confidently tell you that sometimes we're wrong right to begin to see that from models. It's a little bit of a nuance but it is a nice thing to call out when you do see it within here. I'll say right about GPD 5.1 and the upcoming rumors about Gemini 3 as well as some official information around Anthropics Sonnet 4.5. Okay, so I'm going to go ahead and send that in. So next, this is a little bit of an aside, but I do quite like the loading states. They're quite small. They're quite minimal. It doesn't have a bunch of streaming tokens when you don't necessarily need to see all the tokens because quite frankly, oftent times it is quite hard to read all of those different things. So it's gone through, it's performed a couple different searches. And then within here, you can expand to see the MCP search results as well as the different pages and markdown that it scrapes. So it says, "Excellent. I have now the comprehensive information on all three topics. Let me update my progress and write the blog post." Now, one thing that I really like about this is actually its loading stage. It's very nice. It's nice and elegant. It's not actually streaming all of those different tokens where I find a little bit about the streaming tokens effect is oftent times I feel now it's oftent times a demonstration of look how fast the model is or look how much it's doing. But one of the things that I find especially within a sidebar is it will stream out all of those different tokens and you won't actually see what it's doing. And one of the things with all the tokens streaming in, especially within a sidebar, is oftent times it's like a little bit of a performative thing where you just have this sort of feeling like it's doing something or it's really fast or something. Whereas this is a little bit more elegant. And further, what's nice with this is you can turn on the tracking mode and you can actually follow through the different pages. So as it goes and actually edits files, it will put it up on the lefth hand side there and it will go through and focus on all of the different files as we go through those different changes. That is one really nice aspect of that. The last thing that I'm going to say is I want to make sure I have all of these blog post showing on the homepage. Now, additionally within here is you can set the model that you want to use. I think most people are probably going to use sonnet for the most part, but say if you want to have the speed for smaller edits, you can also leverage haiku for that as here we go. So, we have that article that it created for us. So, November 2025, the AI model wars GPT 5.1, set 4.1, and Gemini 3 countdown. We have this really cool blog post that it created for us. We have these nice tables. We have it in the nice sort of cool look and feel of our overall site. And then we also have a number of nice metrics within here. Overall, it actually made a really nice engaging blog post. I'm actually a little bit surprised in terms of the output. And within here, we can see Gemini 3. Gemini 3 is the talk of Discord channels. Here's what we know. And it broke out all of the different information. You can obviously leverage it for web apps, but additionally, if you want to look up documentation or what have you, you can do that as well. But what's nice with this is you can ground it with the search results from the internet and then also be able to generate this very rich blog post. It also has sources and all of that within here as well. Overall, I just wanted to show you a really quick one on zed. And the really nice thing with this is let's say you want to leverage the codec cli and try out the brand new GPD 5.1. You can just simply click that, go through the motions, and overall have that same look and feel for all of the different threads and everything that I had shown you within this cloud code demo. And additionally, let's say Gemini 3 comes out, you'll be able to just simply click that. It will open it up in a new thread. You'll have it within the nice rich editor view within Zed. And not to mention, it's a very performant option. Like I mentioned, overall, I definitely encourage you to check out Zed. It is a great option in terms of code editors. But otherwise, if you found this video useful, please like, comment, share, and subscribe. Otherwise, until the next