Delegation

What is delegation?

Delegation is the act of assigning a bounded task or subgoal to another agent while retaining responsibility for the larger objective. The delegating agent or orchestrator defines the expected result, provides relevant context, and later reviews or integrates the returned work.

Delegation is useful when specialization, isolation, or parallel execution improves the task.

How delegation works

A report-writing agent may delegate competitor research to a sub-agent. A strong assignment includes:

• a specific objective;
• required evidence and output format;
• relevant context and constraints;
• allowed tools and permissions;
• time, cost, and retry limits;
• escalation behavior when blocked.

The parent remains responsible for checking whether the result supports the overall report.

Delegation versus handoff

Delegation gives another agent a part of the work while the delegator keeps ownership of the larger task. An agent handoff transfers active responsibility for the task or conversation.

After delegation, the result normally returns to the delegator. After a handoff, the receiving agent becomes the current owner.

Delegation versus tool calling

A tool performs a defined capability when called. A delegated sub-agent can reason, use several tools, react to observations, and adapt its approach within a bounded goal.

Not every task needs agentic delegation. A deterministic calculation, lookup, or transformation is often better handled as a tool.

Common failure modes

Vague assignments produce incompatible or incomplete results. Too much context increases cost and privacy risk; too little context causes incorrect assumptions. Parallel delegates may duplicate work or update shared state inconsistently.

The delegator may also accept a confident result without verifying evidence. Provenance and explicit completion criteria make integration safer.

Why it matters

Delegation lets a system distribute meaningful work across specialized agents and can reduce total time when tasks are independent. It also creates coordination overhead, extra model calls, and new failure boundaries.

The right question is whether the delegated unit benefits from an independent agent loop. If not, a normal workflow step or tool is simpler. See [Single-Agent vs Multi-Agent Systems](/single-agent-vs-multi-agent-systems/) for delegation trade-offs.