Action-First AI Output Planner
Test one agent, one user, and one repeated task.
1. User preference
What makes an AI response easier to begin?
☐ Action first
☐ Short sections
☐ Numbered steps
☐ Visible progress
☐ Detailed explanation on request
2. Test task
Agent:
Repeated task:
Safe test project:
3. Default output
Time to find the first action:
What caused friction?
4. Action-first output
Time to find the first action:
What improved?
Ten-Rule Review
☐ 1. The first line is an action or answer.
☐ 2. Each numbered step contains one bounded action.
☐ 3. The response ends with one next action.
☐ 4. Tangents are saved instead of interrupting the issue.
☐ 5. Progress is restated visibly.
☐ 6. Time estimates use concrete units.
☐ 7. Completed work is named clearly.
☐ 8. Errors include location, cause, and fix.
☐ 9. Lists remain short or are divided into stages.
☐ 10. Filler and repetitive closers are removed.
Safety and choice
☐ Destructive actions still require approval.
☐ The user can ask for more explanation.
☐ The mode can be disabled.
☐ No diagnosis is stored or inferred.
☐ The user says the format helps.
Decision
☐ Keep on demand
☐ Customize and retest
☐ Make persistent for this user/project
☐ Remove
One improvement:
Next test:
Support the person. Do not define the person.