Comparison
Ordewell vs Claude Code agent teams and subagents
Checked on , against Anthropic's Claude Code documentation and license. Every statement about Claude Code links to its source.
Claude Code can split work across several Claude sessions by itself. Ordewell turns a goal into a plan whose tasks can go to different agents, and Claude Code is one of them. The two overlap, and they combine.
Disclosure: the people who build Ordewell wrote this page. Where a difference could not be sourced from Claude Code's own pages, it is left out.
What each one is for
In Anthropic's words, Claude Code "is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools" (overview). It has two ways to spread work:
- Subagents take side tasks: "the subagent does that work in its own context and returns only the summary" (subagents).
- Agent teams "let you coordinate multiple Claude Code instances working together. One session acts as the team lead, coordinating work, assigning tasks, and synthesizing results." They are experimental and off by default;
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1turns them on (agent teams).
Ordewell's README puts it this way: "One goal becomes a plan graph you can edit, and each task gets its own agent, model and effort." Independent tasks run in parallel on their own branches, pass only on evidence, then merge (README).
How they differ
On the things Ordewell is built around. Each cell about Claude Code links to its source.
| Ordewell | Claude Code | |
|---|---|---|
| Who plans | A planner reads your repository without changing it, asks what is vague and writes tasks with dependencies. It can be a coding agent you are signed in to, or an API provider (README). | The lead session: you describe the work in natural language and it assigns and delegates tasks (agent teams). Claude hands a task to a subagent when it matches that subagent's description (subagents). |
| Plan editable before anything runs | Yes. Change any prompt, dependency, agent, model, effort or mode, then approve it; the desktop app shows the plan as a list or a graph. Build tasks wait for that approval unless you turn on autorun (README). | The lead creates tasks in a shared task list; a task can depend on others and stays blocked until they complete (agent teams). A teammate can be made to plan first and send its plan to the lead for approval (agent teams). |
| Agent and model per task | Each task names its own agent, model and effort. Claude Code, Codex, OpenCode, Pi, GitHub Copilot CLI, Qwen Code, Cline, Gemini CLI and Cursor mix in one plan (README). | Every teammate is a Claude Code instance. The spawn prompt, or a subagent definition's model, picks each teammate's model (agent teams). |
| How a task counts as done | The agent reports through task_complete, or a script task exits cleanly. A clean exit with no report is a failure, and no second model decides (README). |
Teammates mark tasks completed. A TaskCompleted hook can refuse completion by exiting with code 2 (agent teams). |
| Isolation and merging | With git, each task runs on its own branch in its own worktree. You review the diff and merge, or turn on automatic merging. Without git, tasks share your folder (README). | A subagent with isolation: worktree runs in a temporary git worktree, removed if it makes no changes (subagents). For teams, the docs advise giving each teammate its own set of files (agent teams). |
| Surfaces | Desktop app, terminal UI, VS Code extension, CLI and a local API, on macOS, Linux and Windows (README). | The terminal, IDE extensions, a desktop app and the web (overview). |
| License | Apache-2.0 (LICENSE). | Proprietary: "All rights reserved. Use is subject to Anthropic's Commercial Terms of Service" (LICENSE.md). |
When Claude Code is the better choice
- You work in Claude Code alone: subagents and teams live inside the session you already have.
- The work is research or review where teammates "share and challenge each other's findings", or debugging "with competing hypotheses", two of the uses Anthropic lists first (agent teams).
- You want teammates to message each other directly, and to talk to any one of them yourself without going through the lead (agent teams).
- A side task would fill your conversation with search results or logs and you only need the summary. That is what subagents are for (subagents).
When Ordewell is
- You want the plan as something to read and change before build work starts: every prompt, dependency, agent, model, effort and mode.
- You want Claude Code next to other vendors' agents in one plan, such as Codex on one task and OpenCode on another, each with its own model and effort. Claude Code is one of the agents Ordewell runs.
- You want completion to rest on evidence without writing a hook: a task passes when its agent reports through
task_completeor its script exits cleanly. - You want each task on its own branch and worktree, then a diff to review and merge, or automatic merging once you turn it on.
Try Ordewell
Ordewell is open source under Apache-2.0 and runs on macOS, Linux and Windows. Get the desktop app from the download page, or install the terminal app:
# macOS and Linux: a standalone binary, no Node.js
curl -fsSL https://ordewell.ai/install.sh | sh
# Homebrew
brew install ordewell/tap/ordewell
# npm, Node 20.3+: the way to install on Windows
npm install -g ordewell
On Windows the terminal UI runs natively; its mouse and paste need Node 22.17+. git is
optional, 2.38+ recommended: with it, tasks run in their own worktrees. You also need a
supported coding agent, installed and signed in, or a provider API key. Then run
ordewell in your project and type a goal. The docs
cover the rest.
Sources
Every page below was read on 21 November 2026.