Pattern: Reflection
Pattern: Reflection
Category: Reasoning Source: FOR-0012 Status: Documented
When to Use
When an agent needs to self-assess the quality of its own output before delivering it. Unlike the evaluator-optimizer loop, reflection is a single pass of self-critique — generate, critique, revise — rather than an iterative loop with scoring.
How It Works
- The agent generates a first draft
- A critic (the same agent in a different mode, or a separate critic agent) reviews the draft against quality standards
- The critic produces structured feedback: what is good, what needs improvement, specific suggestions
- The original agent revises based on the feedback
- The revised output is delivered (optionally through another review pass)
Example
A digital talent drafting a client proposal generates an initial version, then switches to critic mode: "Does this proposal address all stated requirements? Is the tone appropriate? Are there factual claims that need verification?" The critic flags two issues, and the agent revises before delivering.
Tradeoffs
| Pro | Con |
|---|---|
| Catches obvious errors before delivery | Adds latency (two passes minimum) |
| Works with a single agent — no multi-agent setup needed | Agent may have blind spots in self-critique |
| Improves quality for creative and reasoning tasks | Double the token cost per output |
| Can be structured with explicit checklists | Not a substitute for external validation |
Factory Usage
Not yet used as an explicit standalone pattern, but the concept underpins the Role Factory's TEST stage where generated artifacts are evaluated against checklists before proceeding. Future digital talents could use reflection for any content-generation task.