Confidence-Checking Workflow
Always ask the agent, for each part of its output, to state its confidence and identify which parts need human verification, like triaging a junior analyst's work.
Problem
A homogeneous output hides per-part confidence variation. The user has no signal to apply expertise selectively. The agent has the information (it 'knows' where it is uncertain) but defaults to confident prose throughout.
Solution
Modify the agent's output template to require per-part annotations: each conclusion / fact / recommendation tagged with confidence (high/medium/low or numeric) and a 'verify' flag for the riskiest parts. The user UI surfaces these annotations prominently. Time saved is spent on the flagged parts, not on full re-verification. Pair with confidence-reporting, false-confidence-syndrome (the failure this addresses), reflexive-metacognitive-agent.
When to use
- Analytical outputs with mixed-confidence parts.
- Users have expertise to apply selectively.
- Workflow can carry per-part annotations through the UI.
Open the full interactive page →
Diagram, neighbourhood map, code examples, related patterns and full provenance.