Zed Code Editor: My Experiences with the New Agent Panel
Do you already know Zed? If not: Zed is a open-source code editor written in Rust that focuses on performance and clean, minimalist design. It resembles Visual Studio Code in appearance and operation. As an open-source project, it has already received a lot of attention in the developer community and is now positioning itself as one of the fastest AI code editors currently available. After a tip from a colleague, I installed Zed and tried out the new "Agent Panel" feature intensively for a few days. Here are my first impressions on Linux (Fedora).
Interacting with the AI agent feels surprisingly natural. I simply type in my question, press Enter, and it's ready to go. I can directly ask questions about my code, request changes, or have new code generated. I was impressed by how quickly the agent understands my project structure. While it works in the background, I can continue with other things - really saves time. After the work is done, Zed shows me all changes in a diff window, where I can still adjust anything I don't like.
What's special is that Zed's "Agent Panel" also relies on the terminal when it needs information. For example, if I talk about my PHP version but don't tell the agent which one, it simply runs "php -i" to get the current version. Or if I ask whether PHP's Opcache is enabled, it simply searches for it (php -i | grep -i opcache) and gives me the corresponding answer. If the opcache isn't active, the agent can activate it. It can also create or read files in a directory. For me, this is the first real developer assistant. Of course, it always asks before executing terminal actions or files. This confirmation can be turned on or off.
The help with navigating unfamiliar code is also practical. Instead of wading through unknown structures, the agent finds the important spots and I maintain an overview. Especially with larger projects, this makes a noticeable difference.
Since I work professionally with sensitive code, I always pay attention to data privacy. I like Zed's approach here. According to the developer team, my conversations with the agent are not stored or used for training purposes by default - nothing wanders onto servers without permission. If I want to give feedback, I can use the thumbs buttons, but that's completely optional. I appreciate that confirmation is requested before potentially risky actions like terminal commands. So nothing happens that I don't want.
What I also like: I can decide which AI model I want to use. Claude 3.7 Sonnet, Gemini 2.5, or even custom models via Ollama - everything is possible. Depending on what I'm working on, I set the agent's permissions. Sometimes it needs full access to make changes, sometimes it should only be able to read. I find the extensibility through the new Model Context Protocol (MCP) exciting, as it allows the agent to be used for database access or pull requests. There's still a lot of potential there.
The good news: Zed remains completely free without AI functions (according to the developer, forever). For the AI features, there are different options: I can integrate my own API keys, run models locally via Ollama, or subscribe. The free plan includes 50 prompts per month – enough for me to try it out. Those who need more can take the Pro plan with 500 prompts for $20/month.
After a few days with the Agent Panel, I'm thrilled. It makes my work more efficient and takes some tedious tasks off my hands. The combination of speed, AI integration, and the fact that everything is available as open source under the GPL-3 license has convinced me. I now use ZED much more than before (previously I exclusively used Jetbrains IDEs, which are also very good).
For anyone who likes to code with AI support, taking a look at Zed is definitely worth it.
Zed is currently available as a stable version for Linux and macOS. If you use Windows – the Windows version is still in beta, but according to the website, it can already be tested.
Regards,
Frank
The Agent Panel: My New Coding Partner
Interacting with the AI agent feels surprisingly natural. I simply type in my question, press Enter, and it's ready to go. I can directly ask questions about my code, request changes, or have new code generated. I was impressed by how quickly the agent understands my project structure. While it works in the background, I can continue with other things - really saves time. After the work is done, Zed shows me all changes in a diff window, where I can still adjust anything I don't like.
What's special is that Zed's "Agent Panel" also relies on the terminal when it needs information. For example, if I talk about my PHP version but don't tell the agent which one, it simply runs "php -i" to get the current version. Or if I ask whether PHP's Opcache is enabled, it simply searches for it (php -i | grep -i opcache) and gives me the corresponding answer. If the opcache isn't active, the agent can activate it. It can also create or read files in a directory. For me, this is the first real developer assistant. Of course, it always asks before executing terminal actions or files. This confirmation can be turned on or off.
The help with navigating unfamiliar code is also practical. Instead of wading through unknown structures, the agent finds the important spots and I maintain an overview. Especially with larger projects, this makes a noticeable difference.
Data Privacy? Yes, That's Considered Too
Since I work professionally with sensitive code, I always pay attention to data privacy. I like Zed's approach here. According to the developer team, my conversations with the agent are not stored or used for training purposes by default - nothing wanders onto servers without permission. If I want to give feedback, I can use the thumbs buttons, but that's completely optional. I appreciate that confirmation is requested before potentially risky actions like terminal commands. So nothing happens that I don't want.
Selectable AI Models and Permissions
What I also like: I can decide which AI model I want to use. Claude 3.7 Sonnet, Gemini 2.5, or even custom models via Ollama - everything is possible. Depending on what I'm working on, I set the agent's permissions. Sometimes it needs full access to make changes, sometimes it should only be able to read. I find the extensibility through the new Model Context Protocol (MCP) exciting, as it allows the agent to be used for database access or pull requests. There's still a lot of potential there.
What Does It Cost?
The good news: Zed remains completely free without AI functions (according to the developer, forever). For the AI features, there are different options: I can integrate my own API keys, run models locally via Ollama, or subscribe. The free plan includes 50 prompts per month – enough for me to try it out. Those who need more can take the Pro plan with 500 prompts for $20/month.
After a few days with the Agent Panel, I'm thrilled. It makes my work more efficient and takes some tedious tasks off my hands. The combination of speed, AI integration, and the fact that everything is available as open source under the GPL-3 license has convinced me. I now use ZED much more than before (previously I exclusively used Jetbrains IDEs, which are also very good).
For anyone who likes to code with AI support, taking a look at Zed is definitely worth it.
Zed is currently available as a stable version for Linux and macOS. If you use Windows – the Windows version is still in beta, but according to the website, it can already be tested.
Regards,
Frank
Please also mark the comments that contributed to the solution of the article
Content-ID: 671401
Url: https://rootdb.com/report/zed-code-editor-my-experiences-with-the-new-agent-panel-671401.html
Printed on: May 8, 2025 at 20:05 o'clock