[Developer Guide] How to Precisely Control Agent Scope in the Zed Editor

Introduction: New Challenges in the Era of Agentic Engineering

Recent trends in software development are rapidly evolving beyond simple code generation toward the era of "Agentic Engineering," where AI can independently reason and execute tasks. While the past was characterized by developers copying and pasting AI-generated code, we are now transitioning into a role where developers act as supervisors, delegating specific tasks to AI agents.

However, as the utility of agents increases, a new problem arises: "Agent Sprawl." As multiple AI agents operate simultaneously within a project, it becomes increasingly difficult to control their access levels and monitor which files they are modifying. Unrestricted permissions can compromise code consistency and lead to unexpected side effects.

In this context, recent updates to the Zed editor are highly significant. Nathan Sobo, co-founder and CEO of Zed, emphasized that "a software engineer's contribution should be measured not by the number of lines of code produced, but by their ability to design reliable systems that are easy to change and pleasant to use." Moving beyond mere AI tool usage, Zed provides a core control mechanism for agentic engineering—combining human craftsmanship with powerful AI tools to build better software.

Threads Sidebar: Granular Control Over Agent Permissions

Zed's newly introduced Threads Sidebar serves as the ultimate control tower for determining an agent's scope of work. While agents previously often held vague permissions across entire projects, developers can now use this sidebar to precisely manage which folders and repositories an agent can access.

The greatest advantage of the Threads Sidebar is "project-based grouping." By grouping all active threads by project, you can monitor at a glance which agent is focused on which task. This plays a critical role in preventing chaos in multi-agent environments.

Technically, the most noteworthy feature is Worktree Isolation. Developers can restrict an agent's scope to a specific thread or isolate certain worktrees to ensure an agent's influence does not extend to them. This allows for sophisticated operations—such as protecting folders containing critical core logic while granting full modification rights to agents performing experimental feature implementations.

Multi-Agent Orchestration: Parallel Tasks and Efficient Management

Zed's new features focus on "Orchestration"—going beyond the use of a single agent to managing and running multiple agents in parallel within the same window. According to Zed's blog (Introducing Parallel Agents in Zed), developers can now manage complex workflows by running several agents simultaneously within a single window.

The key strategy here is "Mix and Match." There is no need to use the same AI model for every thread. You can strategically match threads with different models: using lightweight agents for simple refactoring or documentation, and high-performance agents for complex logic design. This approach allows you to optimize both cost-efficiency and execution speed.

Another powerful capability is "Cross-repo work." You can configure a single agent thread to perform read and write operations across multiple repositories. For example, in a scenario where a client library needs to be automatically updated based on changes in an API server, a developer can complete the integrated task through a single agent thread without needing to open separate windows.

UI Optimization and Workflow Tips for Practitioners

The first thing you must adapt to when transitioning to an agent-centric environment is the changed panel layout. Recognizing the importance of agent workflows, Zed is designed to place the Threads Sidebar on the left (next to the Agent Panel) by default. By placing the Project and Git panels on the right, the IDE is configured so that developers can naturally switch tasks with the agent threads as their central axis.

To maintain efficient organization, you should actively use the sidebar's management features. Use the "Archive" function for completed threads to keep your list clean, and if an error occurs or control is needed, immediately hit the "Stop" button to halt the agent's work. It is also a good habit to quickly create new threads whenever a new task arises.

If you prefer a more familiar layout, you can adjust panel positions through customization. You can change the docking position by right-clicking panel icons in the bottom bar or use the Settings Editor for even more granular UI customization. As the scale of agent-driven work grows, these personalized settings become essential for reducing developer cognitive load.

Conclusion: Zed as a Tool for Building Better Systems

Ultimately, the essence of software engineering is not simply generating large volumes of code. As Zed's philosophy suggests, our goal should be "to create reliable and well-designed systems, rather than just increasing lines of code." The technology to control agent permissions and scope is not merely a means to prevent bugs; it is an essential process for ensuring software quality and stability by keeping powerful AI tools within controllable boundaries.

Built on a foundation of smooth 120fps performance, Zed provides extensibility based on open source. Developers can focus on designing more robust systems through sophisticated agent orchestration, rather than being overwhelmed by the sheer volume of code generated by agents. At the center of this future-oriented development environment—where AI and human craftsmanship harmonize—lies Zed's agent control technology.