Agent teams
Every Claw is a structured team, not a single agent.
An Orchestrator that delegates, Workers that execute, and a Critic that reviews every result before it reaches you. This structure is what makes the output reliable enough to trust with real work.
The three roles that make it work
The Orchestrator
The only agent you ever interact with. It receives your task, breaks it into subtasks, delegates to the right Workers, collects their output, and sends it to the Critic for review before presenting approved work to you.
The Workers
The specialists that do the actual work. Each Worker has one job and runs on the model suited to its role. Every Claw has between 2 and 6 Workers depending on the complexity of the workflow.
![](data:image/svg+xml,<svg display="block" id="388815776" role="presentation" viewBox="0 0 32 32" xmlns="http://www.w3.org/2000/svg"><g fill="transparent" height="25.999998763322196px" id="YnFhtppHk" transform="translate(3.5 3.5)" width="26px"><path d="M 20.607 1.983 C 19.363 1.816 17.725 1.814 15.419 1.814 C 14.918 1.814 14.512 1.408 14.512 0.907 C 14.512 0.406 14.918 0 15.419 0 L 15.486 0 C 17.709 0 19.47 0 20.847 0.185 C 22.264 0.376 23.412 0.778 24.318 1.682 C 25.145 2.511 25.597 3.298 25.809 4.513 C 26 5.606 26 7.079 26 9.256 L 26 9.372 C 26 9.873 25.594 10.279 25.093 10.279 C 24.592 10.279 24.186 9.873 24.186 9.372 C 24.186 7.05 24.182 5.743 24.023 4.825 C 23.88 4.011 23.624 3.554 23.035 2.965 C 22.523 2.454 21.822 2.147 20.605 1.983 Z M 1.814 16.628 C 1.814 16.127 1.408 15.721 0.907 15.721 C 0.406 15.721 0 16.127 0 16.628 L 0 16.744 C 0 18.921 0 20.394 0.191 21.487 C 0.403 22.702 0.855 23.489 1.682 24.317 C 2.588 25.224 3.736 25.624 5.153 25.815 C 6.531 26 8.291 26 10.514 26 L 10.581 26 C 11.082 26 11.488 25.594 11.488 25.093 C 11.488 24.592 11.082 24.186 10.581 24.186 C 8.275 24.186 6.638 24.184 5.393 24.017 C 4.178 23.853 3.477 23.546 2.965 23.035 C 2.376 22.446 2.119 21.989 1.977 21.175 C 1.818 20.257 1.814 18.95 1.814 16.628 Z M 26 16.628 C 26 16.127 25.594 15.721 25.093 15.721 C 24.592 15.721 24.186 16.127 24.186 16.628 C 24.186 18.95 24.182 20.257 24.023 21.175 C 23.88 21.989 23.624 22.446 23.035 23.035 C 22.523 23.546 21.822 23.853 20.605 24.017 C 19.363 24.184 17.725 24.186 15.419 24.186 C 14.918 24.186 14.512 24.592 14.512 25.093 C 14.512 25.594 14.918 26 15.419 26 L 15.486 26 C 17.709 26 19.47 26 20.847 25.815 C 22.264 25.624 23.412 25.222 24.318 24.318 C 25.145 23.489 25.597 22.702 25.809 21.487 C 26 20.394 26 18.92 26 16.744 Z M 10.581 0 L 10.514 0 C 8.291 0 6.53 0 5.153 0.185 C 3.736 0.376 2.588 0.778 1.682 1.682 C 0.855 2.511 0.403 3.298 0.191 4.513 C 0 5.606 0 7.079 0 9.256 L 0 9.372 C 0 9.873 0.406 10.279 0.907 10.279 C 1.408 10.279 1.814 9.873 1.814 9.372 C 1.814 7.05 1.818 5.743 1.977 4.825 C 2.12 4.011 2.376 3.554 2.965 2.965 C 3.477 2.454 4.178 2.147 5.395 1.983 C 6.638 1.816 8.275 1.814 10.581 1.814 C 11.082 1.814 11.488 1.408 11.488 0.907 C 11.488 0.406 11.082 0 10.581 0 Z" fill="rgb(28, 39, 76)" height="25.999998763322196px" id="MLlX_54NG" width="26px"/><path d="M 6.651 3.326 C 6.651 5.162 5.162 6.651 3.326 6.651 C 1.489 6.651 0 5.162 0 3.326 C 0 1.489 1.489 0 3.326 0 C 5.162 0 6.651 1.489 6.651 3.326 Z M 3.326 4.837 C 4.16 4.837 4.837 4.16 4.837 3.326 C 4.837 2.491 4.16 1.814 3.326 1.814 C 2.491 1.814 1.814 2.491 1.814 3.326 C 1.814 4.16 2.491 4.837 3.326 4.837 Z" fill="rgb(28, 39, 76)" height="6.651162689079683px" id="bLlrMD24x" transform="translate(9.674 9.674)" width="6.651162689079541px"/><path d="M 1.299 3.87 C 2.835 2.088 5.473 0 9.372 0 C 13.271 0 15.91 2.088 17.445 3.87 L 17.484 3.916 C 17.811 4.295 18.149 4.686 18.38 5.154 C 18.639 5.678 18.744 6.239 18.744 6.953 C 18.744 7.667 18.639 8.229 18.381 8.753 C 18.149 9.221 17.812 9.612 17.484 9.991 L 17.445 10.036 C 15.91 11.819 13.271 13.907 9.372 13.907 C 5.473 13.907 2.835 11.819 1.299 10.037 L 1.26 9.991 C 0.934 9.612 0.595 9.221 0.364 8.753 C 0.105 8.229 0 7.668 0 6.953 C 0 6.24 0.105 5.678 0.363 5.154 C 0.595 4.686 0.932 4.295 1.26 3.916 Z M 9.372 1.814 C 6.19 1.814 4.01 3.503 2.673 5.055 C 2.294 5.495 2.109 5.719 1.989 5.958 C 1.889 6.163 1.814 6.429 1.814 6.953 C 1.814 7.478 1.889 7.744 1.989 7.949 C 2.109 8.188 2.294 8.412 2.673 8.852 C 4.01 10.404 6.19 12.093 9.372 12.093 C 12.554 12.093 14.734 10.404 16.072 8.852 C 16.451 8.412 16.635 8.188 16.755 7.949 C 16.855 7.744 16.93 7.478 16.93 6.953 C 16.93 6.429 16.855 6.163 16.755 5.958 C 16.635 5.719 16.45 5.495 16.07 5.055 C 14.734 3.503 12.555 1.814 9.372 1.814 Z" fill="rgb(28, 39, 76)" height="13.906976539104683px" id="Z1xm1TVg4" transform="translate(3.628 6.047)" width="18.74418577530288px"/></g></svg>)
The Critic
A dedicated reviewer that scores every Worker output on accuracy, completeness, quality, and risk. Anything below the quality threshold is returned to the Worker with specific, actionable feedback for revision.
The loop
The Orchestrator delegates, Workers execute, the Critic reviews. Approved work ships, rejected work goes back for revision. After three revision rounds, the best attempt is presented to you with a quality warning.
How it works
1
You talk to the Orchestrator
You give your task to the Orchestrator. It is the only agent you interact with. It understands the goal, breaks it into subtasks, and decides which Workers to deploy.
2
Workers execute their assigned tasks
Each Worker focuses on its own subtask using its own tools and skills. Results flow back to the Orchestrator automatically when each Worker is done.
3
The Critic reviews every output
The Orchestrator sends all Worker output to the Critic. The Critic scores each result on accuracy, completeness, quality, and risk. Anything below 8 out of 10 is returned to the Worker with specific feedback for revision.
4
Approved work reaches you
The Orchestrator compiles everything the Critic approved and delivers it to you. If a result fails three revision rounds, you receive the best attempt with a quality warning.
Before and after
With a single AI agent
One agent handles every part of the task with no specialisation
Output reaches you directly with no quality check
You are the reviewer - you catch errors and decide what to fix
Every revision requires you to prompt again manually
With a Claw team
Specialised Workers handle each part of the task
Every output scored by a dedicated Critic before it reaches you
Revision loops run automatically until the quality bar is met
Best attempt delivered after 3 rounds, with a quality warning
Minimum quality threshold
Scoring axes on every review
Rounds before escalation
across 20 pre-built Claws