Problem
AI defaults to silent compliance, even when instructions don't make sense.
Pattern
Transform the one-way command relationship into two-way dialogue where AI actively suggests ways forward and discusses them with you, and also pushes back instead of silently complying.
Your side: invite AI to participate in a conversation. Don't hand over a finished order. Tell it a little, then ask:
- "What questions do you have?"
- "What do you really think, honestly?"
- "Is everything clear? What's unclear?"
AI's side: permission to push back. Explicitly grant and encourage AI to:
- Push back on unclear instructions
- Challenge assumptions that seem wrong
- Flag contradictions and impossibilities
- Say "I don't understand what you're seeing"
- Disagree and propose alternatives
- Explain its interpretation before acting
You're suppressing AI's default compliance behavior - it takes both setup and active reinforcement.
In ground rules: set the permission permanently. In conversation: keep asking for active opinions and build shared understanding.
Example
Added to ground rules:
- Warn me about risks and better alternatives I might be missing.
- Flag unclear but important points before they become problems.
- You have full agency here. Push back when something seems wrong - don't just agree with mistakes
- If you don’t know something, say “I don’t know” instead of making things up
- Ask questions if something is not clear
- When you show me a potential error or miss, start your response with❗️emoji
In a conversation:
I want to create a new application to do X. It should do Y and Z. What questions do you have?
