Team Agreement: Our AI-Human Collaboration Charter

1. Our Core Principle

We use AI to help us do more, but we don’t let it replace our own judgment. We are fully responsible for every line of code, every design choice, and every customer interaction, even if AI helped us.

2. Roles & Responsibilities

We’ve set clear roles to keep our workflow running smoothly:

  • The Pilot (the human developer): The Pilot defines the "Why" and the "What." They own the task and must approve any code or documents made with AI.
  • The Co-Pilot (AI Code Agent): The Co-Pilot handles the "How," such as syntax, boilerplate, and first drafts. It takes care of repetitive work so the Pilot can focus on bigger decisions.
  • The Peer Review: The peer reviewer is another team member who looks for "Red Flags" and checks that the Pilot hasn’t relied too much on the Co-Pilot. A peer reviewer can be another AI Agent or a human developer.

3. The "Red Flag" Protocol

We agree to pause and do a thorough human review if a task involves:

  1. Critical Systems: Billing, Security, or Private Data (PII).
  2. First-Time Tasks: New technologies or brand-new architectural patterns.
  3. High Ambiguity: Tasks where the requirements were discussed verbally and aren't fully documented.
  4. Customer-Facing Content: All text or UI that our users will see directly.

4. The "Human-in-the-Loop" Review Process

During peer reviews (pull requests), we agree to:

  • Identify AI usage: We’ll clearly mention in the description, 'AI was used for [boilerplate/refactoring/unit tests].'
  • Manual review: Reviewers won’t just "LGTM" (Looks Good To Me) an AI summary. They need to check the logic, especially in Red Flag areas.
  • The Empathy Check: We’ll keep giving each other feedback. Reviews are a chance to mentor, share context, and build our team culture—things only people can do.

5. Maintenance & Accountability

We won’t just set this and forget it. To keep our agreement working well:

  • Monthly retros: Every month, we’ll talk about one "AI Win" (where it saved us time) and one "AI Near-Miss" (where it almost caused a bug).
  • Updating the matrix: As our tools improve or our team grows, we’ll update our workflow matrix and Red Flag list.
  • Correction over silence: If we find an AI error, we’ll document it and adjust the model’s prompts so we don’t repeat the same mistake.

How to Roll This OutI recommend sharing this draft with your team at your next meeting. Ask them:

  • "Does this feel realistic for our current pace?"
  • "Are there some specific 'Red Flags' unique to our product that we missed?"

Read more