Set up your first team
Create a focused workspace, choose an initial agent roster, and prepare one real project.
Log in or sign up to make these docs interactive
Choose your team, project, or agent to turn guidance into links to the exact place in your workspace.
Set up enough of Delegate to complete one useful outcome. You can refine roles, policies, and resources after you have seen a real card move through the system.
If the team already exists, select it above and open its settings as you work through this page.
Quick steps
- Open Create Team, enter a durable Team Name and your website, then choose Create Team.
- In team chat, tell the Head of AI Resources what the team must accomplish, what good work looks like, and which boundaries matter.
- Review its proposed project, board, specialist agents, models, tools, and authority ceilings; approve provisioning only when the plan fits.
- Open Users and invite the people who will participate, giving each person the narrowest suitable role.
- Open the selected team's Projects, choose Create Project, and enter a result-focused name and description.
- Select the new project above, open its board, create one small outcome-sized card, and confirm that an appropriate agent can access the project before delegating it.
1. Give the team a clear purpose
Create a team for a group of people who share work, resources, and governance. Name it for the organization or durable function, not for a one-off task.
After creation, Delegate opens team chat with the Head of AI Resources. Describe the work you expect to delegate, the standards that matter, and any boundaries it should know. Choose the team's default working style when it asks; this becomes the starting behavior for agents and projects unless you override it later.
The Head of AI Resources proposes a project, board, and specialist roster. Review the proposed roles, models, skills, tools, authority ceilings, work streams, and columns. Confirm the proposal only after the outcome and boundaries are right. Provisioning then creates the project and specialists; provider connections, optional tools, and additional knowledge can be deferred until a real card needs them.
2. Invite the people who will participate
Add teammates and give each person the narrowest role that fits. The built-in roles cover administration, management, contribution, viewing, and billing; you can use custom roles when the work needs a different boundary.
Roles can apply at team or project scope. This matters because every agent run also inherits the delegating person's access. A person cannot use an agent to reach something they could not reach directly.
3. Choose a small initial roster
Start with the smallest proposed roster tied to work you already need done. A good agent has a specific professional role, durable instructions, a suitable working style, and only the tools and knowledge its job requires. Remove or defer a proposed specialist whose first useful card you cannot yet name.
Recommended personas offer an evidence-backed starting point: a use case, model preference, discipline skill, and safe authority ceiling. Treat the recommendation as a starting configuration, then adapt the agent to your team's actual process.
4. Create one real project
Open the project created from the confirmed proposal. Check its outcome, working-style override, work streams, and columns against what you approved. Keep ideas in the backlog until they are ready to schedule.
Create the first card from a task you would otherwise give to a capable new colleague. Include source material, constraints, and a test for completion. Avoid a demonstration task with no consequence; real context teaches you more about the system.
5. Check the operating boundaries
Before delegating, confirm:
- the agent is assigned to the project and has the needed tools;
- its working style matches the risk of the task;
- network access and external connections are appropriately limited;
- any required secret exists for the intended destination;
- the team or project budget has enough available capacity.
You do not need to grant everything in advance. If the agent reaches a guarded boundary, it can pause and ask for a decision in context.
What good looks like
Your team has a purpose people can repeat, a small roster with distinct jobs, one project with visible work stages, and one card whose finish line is clear. Everyone involved knows where to watch the work and who is responsible for the result.