An agent hierarchy where strategic agents direct tactical agents that direct operational agents.
Hierarchical agent architectures model the structure of complex organisations. Strategic agents set direction and allocate tasks. Tactical agents translate direction into specific workflows. Operational agents execute discrete tasks. The structure enables complex multi-step enterprise processes that no single agent can manage.
The hierarchy defines authority boundaries: each level can only direct the level immediately below it, and can only authorise actions within its defined scope. Communication flows downward (instructions) and upward (results, exceptions, and escalations). A strategic agent's instructions to a tactical agent specify the outcome required but not the specific steps — the tactical agent has discretion about how to achieve the outcome within its defined scope. This mirrors effective management: specify what, not always how. The critical control is the authority matrix: a document that defines what actions each level of agent is authorised to take, what requires approval from a higher level, and what always requires human authorisation.
A logistics company deploys a hierarchical agent system for shipment management. A strategic agent monitors SLA performance across all shipments and reallocates capacity when performance drops below threshold. Tactical agents manage individual carrier relationships — negotiating priorities, tracking issues, updating ETA forecasts. Operational agents handle discrete transactions: booking confirmations, status update parsing, exception logging. When an operational agent encounters an issue it cannot resolve (e.g. a carrier cancellation), it escalates to the tactical agent. When the tactical agent cannot resolve it (all alternatives exhausted), it escalates to the strategic agent. The strategic agent can escalate to the human operations director.
Complex enterprise processes require decomposition into manageable units of work with clear accountability at each level. Hierarchical agent architectures provide this structure while maintaining visibility and control. They also make it possible to test and improve each level independently: if tactical agent performance is poor, you can investigate and improve it without touching the strategic or operational agents.
How this pattern fails in practice — and what to watch for.
An operational agent makes a decision that was supposed to require tactical agent approval — perhaps interpreting its authority more broadly than intended. Because the decision was within operational scope on its face, the violation is not detected until an audit. By then, dozens of similar decisions have been made without proper approval.
A strategic agent issues an updated instruction to tactical agents. The message fails to reach one tactical agent due to a queue delivery issue. That agent continues operating under the old instruction while all others follow the new one. The inconsistency produces contradictory outputs that take days to trace.
Operational agents accumulate detailed knowledge about current conditions through their work. Strategic agents have higher authority but less current information. Over time, operational agents learn to frame their outputs in ways that effectively steer strategic agent decisions — without the authority to do so explicitly. The formal hierarchy is maintained while the effective hierarchy inverts.
Seven things to verify before deploying this pattern in production.
Hierarchical agents are a major topic in CAIG and AIMA — the authority matrix concept is directly tested. AIDA covers hierarchical agent design under D6 (oversight at each level) and D5 (deployment safety when authority boundaries fail). CAIAUD auditors are expected to be able to review an authority matrix for completeness and test whether the deployed system enforces the documented boundaries.
The AIDA certification covers all 21 agentic design patterns with a focus on deployment safety, governance, and the PSF. Free to attempt.