Home/
Part XIII — Expert Mode: Systems, Agents, and Automation/37. Agentic Workflows (Without Letting It Run Wild)/37.3 Planning, executing, and critiquing roles
37.3 Planning, executing, and critiquing roles
Overview and links for this section of the guide.
On this page
Separation of Concerns
Do not ask the same model to write code and check it in the same turn. It has "completion bias"—it tends to think its own output is correct.
Pattern: 1. **Planner:** "Here is the roadmap." 2. **Executor:** "I wrote the code." 3. **Critic:** "I found a bug in line 5. Fix it."
The Critic
The Critic prompt should be mean. "You are a harsh code reviewer. Find security flaws. If none, say 'PASS'."
This adversarial dynamic improves quality significantly.