Skip to main content

Meta Agent

Meta Agent lets you ask one agent to help set up and manage other agents. Instead of configuring every agent by hand, you can describe the business outcome you want, review the proposed changes, and approve the work that should happen.

Use Meta Agent when you want help turning a goal into one agent, a coordinated team of agents, or a cleaner setup for agents you already have.

What Meta Agent Can Help With​

Meta Agent can help you:

  • Create a new agent from a business goal.
  • Create a team of agents with different responsibilities.
  • Connect agents so they can coordinate with each other.
  • Brief agents with roles, responsibilities, and next steps.
  • Review existing agents, including their names, standing instructions, activity state, resource settings, contact setup, files, and recent work history when available.
  • Update names, responsibilities, standing instructions, active or paused state, resource settings, intelligence level, contact rules, and preferred contact routes where those controls are available.
  • Archive agents that are no longer needed.
  • Reorganize responsibilities across an existing set of agents.
  • Add or review small files for an agent when file support is available.
  • Review, add, remove, approve, or reject contact access where contact controls are available.
  • Help shape ongoing work through clear standing instructions and timing hints.

Meta Agent is best for setup and management work. It does not replace the agents doing the business work. For example, Meta Agent can create and brief a sales research team; the sales research agents then do the research.

How To Ask​

Start with the result you want. Meta Agent can fill in a practical structure, but better context usually produces a better proposal.

Useful details include:

  • The business area: recruiting, sales, research, customer support, marketing, operations, finance, or another function.
  • The outcome: create a demo team, clean up existing agents, set up a weekly customer report, brief one agent, archive old agents, or connect two agents.
  • Constraints: one-time setup, no recurring check-ins, keep usage low, use a higher intelligence level for complex research, pause anything old, or ask before changing contacts.
  • Inputs: job roles, target accounts, research topics, support queues, campaign themes, vendor names, or files you want an agent to use.

Example:

Create a demo recruiting team for our customer success hiring push. I want one agent to source candidates, one to screen resumes, and one to prepare interview briefs. This is just setup for now, with no recurring check-ins.

Expected outcome: Meta Agent should propose the agents, their responsibilities, and any links or briefing messages needed. It should not add ongoing schedules because the request is setup-only.

Creating One agent​

Ask Meta Agent for a single agent when one clear role is enough.

Example prompts:

  • "Create an agent that researches target accounts before sales calls. Keep it lightweight and ask me before sending anything externally."
  • "Set up one agent to summarize customer feedback from files I provide."
  • "Make an agent for vendor research. It should compare pricing, risks, and renewal timing."

Expected outcome: Meta Agent should propose the name, role, standing instructions, and any settings that matter. After you approve, it can create the agent and brief it with the starting context.

Creating Teams Of agents​

Teams are useful when a business process has multiple roles. Meta Agent can split the work into focused agents, connect the ones that need to coordinate, and brief each one with its responsibilities.

Recruiting​

Prompt:

Create a recruiting team for a support manager role. I need one agent to find candidates, one to screen resumes, and one to prepare interview packets. Do not set up recurring outreach yet.

Expected outcome: A proposed team with separate sourcing, screening, and interview-briefing responsibilities. Links may be proposed where coordination helps. No recurring schedule should be added unless you ask for one.

Sales​

Prompt:

Build a sales prep team for the Northeast territory. One agent should research target accounts, one should clean up account notes, and one should draft call prep briefs.

Expected outcome: A sales-focused team with clear handoffs. Meta Agent may propose links so the account researcher and call prep agent can coordinate, then ask for approval before creating or connecting anything.

Research​

Prompt:

Set up a research team for the electric vehicle charging market. I want one agent for competitors, one for regulations, and one for customer segments.

Expected outcome: A research team with separate charters for each topic. If you did not ask for recurring monitoring, the initial setup should stay one-time.

Customer Operations​

Prompt:

Create a customer ops team to help us organize support escalations, summarize customer themes, and prepare weekly leadership updates.

Expected outcome: Meta Agent may propose escalation triage, theme analysis, and leadership update agents. Because "weekly leadership updates" is explicit timing, the proposal should make that recurring update visible in the approval plan before any schedule-related behavior is added.

Marketing And Content​

Prompt:

Create a content team for our product launch. One agent should collect customer proof points, one should draft post ideas, and one should review the launch calendar.

Expected outcome: A launch-focused team with responsibilities and briefings. Unless you ask for scheduled content ideas or ongoing calendar checks, the team should be set up without a new recurring cadence.

Business Operations​

Prompt:

Create an operations team for vendor renewals. I need help tracking vendor details, preparing renewal questions, and keeping leadership briefed monthly.

Expected outcome: Meta Agent can propose vendor tracking, renewal prep, and leadership briefing agents. Since "monthly" is explicit, any ongoing briefing schedule should be included plainly in the proposal and approval scope.

Connecting And Briefing agents​

Meta Agent can connect agents that need to coordinate. Connected agents can share updates within the limits you approve, which is useful when one agent produces input another agent needs. For the manual setup path and usage examples, see Peer Linking.

Examples:

  • Connect the candidate sourcing agent to the resume screening agent.
  • Link the competitor research agent to the weekly market brief the platform.
  • Connect the support escalation agent to the customer themes agent.

Meta Agent can also brief agents. A brief is a message that gives context, assigns next steps, or explains how an agent should work with another agent.

Briefings use the same peer-DM path as ongoing agent-to-agent coordination. The managing agent must have an approved peer link to the target before it can send the briefing; if that link does not exist, Meta Agent includes it in the proposed changes for approval.

Example:

Brief the account research agent that Acme, Globex, and Northstar are the first target accounts. Tell it to prepare concise sales-call notes and avoid sending external messages.

Expected outcome: Meta Agent should show the briefing action in the approval plan before sending the message.

Managing Existing agents​

Meta Agent can help clean up and maintain the agents you already have.

Common requests:

  • "Show me the agents related to recruiting and summarize what each one does."
  • "Reorganize my sales agents so account research, list cleanup, and call prep are separate responsibilities."
  • "Rename the customer feedback agent to Customer Insights Agent and tighten its instructions."
  • "Pause the old market research agent."
  • "Archive the stale demo agents from last month."
  • "Pause the content ideas agent until the campaign starts."
  • "Use a stronger intelligence level for the legal research agent."
  • "Connect the account notes agent to the call prep agent."
  • "Remove the link between the old sourcing agent and the screening agent."

Expected outcome: Meta Agent can inspect the relevant agents, propose the changes, and ask for approval before making updates.

Charters, Prompts, And Ongoing Work​

Each agent has standing instructions, often called its charter. The charter describes the agent's job, boundaries, and preferred way of working. Your prompts and approvals can update that charter over time.

Some agents also do ongoing work. In normal language, you can steer ongoing behavior by saying what cadence or trigger you want:

  • "This is one-time setup only."
  • "No recurring check-ins."
  • "Send me a weekly brief every Friday."
  • "Monitor this daily and summarize changes."
  • "Check the inbox every morning."
  • "Stop the recurring update for this agent."
  • "Change the monthly vendor review to the first Monday of each month."

Behind the scenes, the platform uses the charter and timing instructions to decide what it should do over time. You do not need to write technical schedule rules. Clear business language is enough.

If you ask for a demo team, a setup-only team, a one-time research pass, a cleanup task, or a project team without ongoing timing, Meta Agent should leave recurring work out of the initial plan. If wording like "monitor," "follow up," or "keep an eye on" could mean ongoing work but you did not give a cadence or clear recurring intent, Meta Agent should ask a clarifying question or leave scheduling out of the proposed change.

Resources, Intelligence, And Usage​

Where supported by your account, Meta Agent can help adjust how much capacity an agent can use.

Examples:

  • "Pause this demo team when setup is complete."
  • "Use a stronger intelligence level for the technical research agent."
  • "Pause the content agent until the campaign starts."
  • "Show me which agents used the most tokens."

Higher intelligence can help with complex reasoning, research, or judgment-heavy work. Review recorded usage for demos, prototypes and background work.

Contacts, Files, And Preferred Contact Routes​

Where available, Meta Agent can help manage how an agent communicates and what supporting material it can use.

Contacts:

  • Review who is allowed to communicate with an agent.
  • Add or remove approved contacts.
  • Approve or reject pending contact requests.
  • Set whether an approved contact can only communicate or can also help configure the agent.

Preferred contact routes:

  • Set or clear the preferred way an agent should reach you when that route is available and safe for your account.

Files:

  • Add small files you provide to an agent's workspace.
  • List files already available to an agent.
  • Brief an agent to use files that are already in its workspace.

Meta Agent should ask before changing contacts, preferred contact routes, or files. It should also avoid exposing unnecessary private contact details in summaries.

Approvals And Safety Boundaries​

Meta Agent can inspect your accessible agents to prepare a proposal, but important changes require your approval.

It should ask before:

  • Creating an agent.
  • Updating an agent's name, charter, active state, settings, resource limits, intelligence level, or ongoing work.
  • Connecting or disconnecting agents.
  • Sending a briefing message to another agent.
  • Uploading a file for an agent.
  • Adding, removing, approving, or rejecting contacts.
  • Setting or clearing preferred contact routes.
  • Archiving an agent.

The approval plan should say what will change. If a proposal includes ongoing work, the timing should be visible in the approval scope. If the request did not ask for ongoing work, the approval plan should not quietly include it.

Meta Agent can only manage agents and settings it is allowed to access in your personal or organization workspace. It should not make changes outside that scope.

Practical Prompt Patterns​

Use these patterns as starting points.

Create one agent:

Create an agent to research enterprise healthcare accounts before sales calls. It should prepare concise account briefs and stay internal-only unless I approve otherwise.

Create a team:

Create a three-agent research team for the warehouse robotics market: competitors, customer segments, and regulatory risk. This is a one-time setup with no recurring check-ins.

Connect and brief:

Connect the competitor research agent to the executive brief agent, then brief the executive brief agent to use competitor updates only as input for weekly leadership summaries.

Reorganize:

Review my customer ops agents and propose a cleaner split of responsibilities for escalations, customer themes, and leadership updates. Do not change anything until I approve.

Adjust resources:

Review inference usage and propose which demo agents should be paused.

Ongoing work:

Update the vendor renewal agent so it prepares a monthly renewal-risk brief for leadership. Include the timing in the approval plan before making changes.

Stop ongoing work:

Stop the recurring check-in for the old product launch agent, but keep the agent available for one-off questions.

Last updated on