Planner-Executor Pattern

Planner-Executor splits agent behavior into two stages: a planner proposes a structured plan, and an executor carries it out via tools.

The plan can be reviewed by policies, humans, or automated checks before any action is taken.

This reduces impulsive actions and improves reliability for multi-step workflows.

Separating planning from execution supports governance: you can inspect intent before impact.

It enables safer automation at scale because you can apply approvals, budgets, and constraints at the plan boundary.

It also improves debugging: failures are tied to specific plan steps, not a single opaque “agent run.”

• Letting the executor silently rewrite the plan (“plan drift”) during execution.
• Plans that are unstructured prose rather than step lists with explicit tools and parameters.
• Skipping plan validation (permissions, data access, budgets) until after execution begins.
• No re-planning strategy when the world changes mid-run.

• Copyl’s tool-first approach aligns naturally with planner-executor: plans map to auditable tool calls.
• Approval workflows can gate execution at the plan or step level for high-risk actions.
• Audit trails provide step-level visibility into what happened and why.

• Tool-Using Agent Pattern
• Human-in-the-Loop (HITL) Pattern
• Retry / Self-Healing Agent Pattern
• Undo / Reversibility Pattern
• View all 20 patterns