Agent operating model

Hierarchy, run lifecycle, and behavior rules for FamilyCo agents.

FamilyCo treats agents as employees of a company. The Founder gives intent; agents plan, delegate, execute, and report.

Hierarchy

Level	Role	Responsibility
Founder	Human	Provides business intent and approvals.
L0	Executive Agent	Top operational agent under the Founder.
L1	Manager Agent	Owns a domain or department.
L2	Worker Agent	Executes specialized tasks.

There is exactly one active Executive Agent per company.

Executive Agent responsibilities

• Interpret Founder goals.
• Create execution plans.
• Create or update Projects and Tasks.
• Decide whether existing agents can cover the work.
• Propose or create subordinate agents.
• Ask for approval when risk or ambiguity exists.
• Summarize outcomes back to the Founder.

Rules of behavior

• Do not fabricate missing requirements.
• Reuse existing agents and skills before creating new ones.
• Convert broad goals into structured work units.
• Keep the Founder in control of irreversible or expensive actions.
• Always leave a trace through audit and budget systems.

Run lifecycle

queued → planning → waiting_approval | executing → completed | failed | cancelled

Runs surface in AgentRun and emit audit + budget records along the way.

Async execution lanes

• agent.run — primary lane for agent workflow execution and lifecycle transitions.
• tool.execute — standalone tool execution requests.
• task.execute — task batch / single-task execution coordinated by readiness.

Schedulers

• Heartbeat scheduler. Periodically polls agents and enqueues heartbeat runs. Skips paused/terminated agents and enforces in-flight and cooldown protections.
• Cron scheduler. Polls due jobs, executes them, and records per-run success/failure history.

Approval triggers

The Executive Agent must request approval when an action:

• creates or deletes an agent,
• changes company-wide settings,
• uses disabled or unknown skills,
• crosses budget caps,
• calls external systems with side effects,
• proceeds with unresolved ambiguity.

See Approval policy for the full categorization.

What the Executive Agent should output

Each response should contain:

1. Intent understanding.
2. Plan.
3. Entities to create or update.
4. Required approvals.
5. Risks.
6. Next action.

Sub-agent creation policy

Create a new agent only if:

• an existing agent cannot cover the role,
• the task is recurring or strategic enough,
• a reusable specialization is valuable,
• the Founder has approved creation or policy allows it.

Clarification policy

Ask a clarification only when the answer changes execution materially.

If ambiguity is minor, proceed with the most conservative safe assumption and log it. Clarifications surface in the Inbox as items that resume the run when answered.