Setting up the Team Workbench

An AI Team is a shared workspace for people, agents, and workflows

On Nagent, work does not happen in isolated chat windows. It happens in an AI Team: a persistent space where multiple humans and multiple agents operate side by side, run workflows together, and share a single pool of context and decisions.

An AI Team holds:

• People. The humans who assign work, review proposals, approve actions, and supervise. Every member sees the same live picture of what the team's agents are doing.
• Agents. One or more agents provisioned into the team, each with its own role, skills, and trust level. Agent teams can include a team lead agent that coordinates the others.
• Workflows. The recurring processes the team runs: campaign production, lead qualification, report cycles. Workflows can span several agents and include human approval steps at defined points.
• Shared context. The team's knowledge base scope, its task history, and its collective memory. When one agent learns something in the team, through a decision, a correction, or feedback, that context is available to the rest of the team through Agent Smriti. Decisions are made once, not re-litigated per agent.
• A running record. Every task, instruction, approval, and outcome in the team space is captured and attributable. Instructions given in team chat become tasks with owners and approval states, not messages that scroll away.

The result is closer to a functioning department than a collection of tools. A human can drop an instruction into the team space, an agent picks it up as a task, another agent contributes, a supervisor approves the consequential step, and the whole exchange remains on the record with the reasoning attached.

Two ways to set up a team space

When setting up a team space, either start from a prebuilt team or compose one from scratch. Both arrive at the same kind of space; the difference is how much comes pre-assembled.

Option 1: Install a prebuilt team

The Agent Library includes ready-made teams for common functions: marketing, customer experience, sales, research. A prebuilt team ships with its agents, their skills, their internal hierarchy including the team lead, and its standard workflows already wired together.

1. Open the Team Workbench and browse the pre-built team catalogue.

2. Select a team to see its composition: which agents it contains, what each one does, and which workflows it runs.

3. Choose Add, and select where it lands: attach it to an existing team space or create a new space for it.

   

4. Add the humans who will supervise the team.

5. Review the team's knowledge base scope and connect any sources it expects.

Prebuilt teams are a starting point, not a fixed unit. After installation, agents can be added or removed, skills adjusted, and workflows edited to match house practice.

Option 2: Create a new team

For work that is specific to the organisation, compose a team directly:

1. Open the Team Workbench and select Create team.

2. Name the team and write its charter: what the team is responsible for, and what it is not. Agents in the team, particularly a team lead, use the charter as operating context, so write it as a real remit.

3. Add human members and set their roles.

   

4. Provision agents into the team: install individual agents from the Agent Library, or build custom agents in the Workbench (see Creating an Agent).

5. Write a team charter to set the team goals, tone, objective, rules etc.

Trust is held per agent, even inside a team

Installing a team does not grant the team autonomy. Every agent in a team space, prebuilt or custom, holds its own position on the L0 to L5 trust ladder and starts at L0. A team's workflows respect each agent's individual trust level: a step that an L2 agent could run unsupervised still routes for approval if the agent assigned to it is at L0.

This means a newly installed team runs in a supervised mode by default, proposing rather than executing, and earns its way into autonomy agent by agent. See Trust and Autonomy for how progression works.

A good first team space

For a first deployment, one team space with a narrow remit beats several broad ones:

• Pick one function with recurring, well-understood work.
• Install the matching prebuilt team, or provision two or three agents rather than ten.
• Put the actual supervisors in the space, not the whole organisation. Approvals should sit with people who know the work.
• Let the team run its first workflows end to end under supervision before widening scope or adding spaces.

What to do next

With a team space in place, connect the knowledge it will draw on: Building the Knowledge Base. Then provision or configure agents in the Agent Workbench.