Zed brings native editing and several coding agents into one workspace

Zed is a code editor built by Zed Industries that combines conventional development tools with AI agents and real-time collaboration. Its appeal is not limited to a single model: developers can use Zed’s own agent, connect an external agent or run a terminal-based workflow. The practical decision is whether that workspace improves the daily loop of understanding code, making a change and reviewing the result.

01 / ProductAn editor with distinct ways to use AI

The AI overview² separates three decisions: how an agent runs, how models are accessed and which editing feature is being used. Zed Agent is the native agent. External Agents connect through the Agent Client Protocol, or ACP. Terminal Threads let developers run an agent’s command-line interface in a terminal-backed conversation. These paths may look adjacent in the interface while using different configuration and accounts.

AI features also range from small edits to full tasks. Edit predictions suggest changes while typing; the inline assistant transforms a selection; the Agent Panel supports reading files, changing code and running commands. The same editor can therefore support a developer writing most code manually or delegating a bounded implementation. That flexibility is useful only when the chosen path remains understandable to the person supervising it.

Zed remains an editor even when AI is disabled. It also offers multiplayer editing and remote projects. A switch should therefore be evaluated on ordinary development work as well as generation: language support, navigation, terminals, debugging and the team’s repository conventions. An impressive agent answer does not compensate for a missing part of the developer’s everyday toolchain.

02 / AudienceWho should consider changing editors

Zed is relevant to developers who want an integrated workspace without tying every AI interaction to one provider relationship. A team may already have approved model endpoints or subscriptions. An individual may want to compare a native agent with an external one while retaining the same editing and review environment. These are concrete reasons to evaluate Zed’s approach.

It can also fit engineers who collaborate closely on code. Real-time shared editing lets two people inspect and change the same project instead of passing screenshots or patches back and forth. This is especially useful for explaining an unfamiliar module or working through a difficult failure with another engineer. The collaboration should still have an agreed repository and branch state.

Compare the Cursor blueprint when evaluating another editor-centered AI workflow. The Replit blueprint is a different comparison when the team also needs a managed environment and application deployment. Zed’s remote development capability connects to machines you can access; it should not be assumed to supply the same hosting and operational package as an application platform.

03 / WorkflowA proposed workflow for repairing a command-line tool

Consider a developer fixing a command-line utility that mishandles filenames containing spaces. This is a proposed evaluation, not a completed test. The change is small enough to inspect, but it crosses argument parsing, filesystem access and user-facing errors. Open the repository in Zed and reproduce the failure using its existing task or terminal setup before introducing an agent.

Choose one agent path explicitly. With Zed Agent, select the configured model provider. With an External Agent, complete that agent’s authentication flow. The external-agent guide³ explains that the external process usually owns its runtime, model selection, authentication and native configuration. A model key saved for Zed Agent does not automatically configure every external agent in the panel.

Ask the agent to find where the filename changes between user input and the filesystem call. Give it the failing command and the expected output, and require a regression test using the project’s established test framework. Avoid requesting a wholesale parser replacement unless the investigation justifies it. The useful result is a small patch with an explanation of the failure, including what happens when the file does not exist.

Review the edits through the Agent Panel⁴ and run the test suite. Check the actual changed files, including any dependency or configuration updates. If the agent takes an unhelpful direction, use the available checkpoint or Git workflow appropriate to that integration. Checkpoint, thread-history and usage features vary across external agents, so verify them with the chosen agent before relying on a familiar-looking control.

Then invite a colleague to inspect the behavior using collaboration⁶. Have that person try a quoted filename, an empty argument and a path with non-ASCII characters. The goal is to see whether collaboration and agent review make the change easier to understand. Retain the final diff and the tests that established the fix. That is a better switching decision than a comparison of how quickly two editors generate a first draft.

04 / PricingPricing separates the editor from hosted model use

Business is described as an organization administration layer, and its members can hold individual Pro subscriptions. It should not be assumed that the higher seat price includes every person’s model consumption. Model policy, data-sharing settings and consolidated visibility are the reasons to buy that layer. Work out how hosted usage and existing provider arrangements will be paid in the actual organization.

The distinction matters when comparing editors. One plan may include a large model allowance, while another offers a lower editor fee and usage-based inference. A third route may use an existing external-agent subscription. Add the costs actually attributable to the chosen workflow and avoid counting the same provider subscription twice when it serves several tools.

For a trial, set a spending limit where the selected route supports one and record the work completed within it. A long investigation can consume more model tokens than a straightforward edit even when the final patch is shorter. Measure the complete fix, review effort and normal editing experience. The number of suggestions accepted alone does not show whether an engineer delivered better work.

05 / DistinctionsWhat stands out in the workspace design

The agent interface can host several kinds of work while the editor remains the place where code is inspected. That reduces the need to move between a separate chat window, terminal and diff viewer. It also makes the configuration boundary more important: two threads may use different tools, instructions and providers despite appearing in the same sidebar.

Remote development⁵ keeps the editor interface local while the source code, language servers, tasks and terminal run on a remote machine over SSH. This is useful when a project needs a particular operating system or a large build environment. It allows a developer to keep a responsive local workspace without moving every build dependency onto a laptop.

The remote split is worth understanding in detail. Zed’s documentation describes local model communication and separate local, server and project settings. External-agent credentials can follow their own arrangement. When an agent cannot reach a tool or provider, identify which process is running where before changing configuration. A clear map of the editor, remote host and agent resolves many apparent integration mysteries.

06 / QuestionsQuestions to answer before a team rollout

Does the chosen language and extension setup cover the team’s work? Try the debugger, formatter, test task and navigation on a representative project. Include a task involving generated code or a less common build system if those are ordinary parts of development. The correct evaluation is the team’s actual workday, not the set of features visible in an introductory demonstration.

Which controls apply to which agent? Zed’s own profiles and skills do not automatically govern an external process. The external-agent documentation sets out where native configuration, forwarded MCP tools and permission handling may differ. Write down the intended agent and provider route for the team so that a developer does not accidentally use a different account or instruction set when opening a new thread.

Business administrators should also check currently shipped identity features. The pricing FAQ states that SSO, SAML and SCIM are planned rather than available. It describes organization controls as applying to Zed’s hosted services. If those identity integrations are a prerequisite, a roadmap statement does not meet it. A smaller pilot can still assess the editor while the organization evaluates its required rollout conditions.

07 / DecisionChoose Zed on the strength of a complete workday

Zed is a compelling candidate when a team values a native editing experience, collaboration and the freedom to choose an agent or model path. Its flexibility rewards a deliberate configuration: one understood route is more useful than a collection of partially configured providers. Begin with the workflow developers already use most often.

Complete a real fix, a review with a colleague and a remote session if remote work matters. Record missing capabilities alongside improvements in navigation and agent interaction. Those observations provide an evidence-based reason to switch or stay.