Agent Design Canvas
Eight questions to answer before you build anything.
1PURPOSE
What job is this agent hired to do?
As a [user], I need to [job], so that [outcome]. What is it explicitly NOT for?
2TRIGGERS
What starts the agent working?
User request? Scheduled event? System alert? Another agent? Which is primary?
3TOOLS & CAPABILITIES
What can it connect to and use?
List each tool. Risk-rate: read-only (low), write/modify (med), irreversible (high).
4KNOWLEDGE & CONTEXT
What does it know and remember?
Instructions, retrieved docs, conversation history, persistent memory. Who controls what it sees?
5AUTHORITY LEVEL
What level of autonomy does it have?
Observe · Recommend · Act w/ Approval · Act & Report · Autonomous (within scope).
6GUARDRAILS
What is it explicitly prohibited from doing?
Hard boundaries (code) vs soft (prompt). Spending limits? Data restrictions? Access controls?
7HUMAN CHECKPOINTS
Where must a person approve or intervene?
HITL (approve each action) · HOTL (monitor) · HOOTL (audit only).
8SUCCESS METRICS
How do you know it's working?
Task completion rate? Accuracy? Cost per run? User satisfaction? What does 'good' look like?
Tip: if you can't fill sections 5–7 with confidence, you're not ready to build. Go back to section 1.