Autonomy Spectrum
What is Autonomy Spectrum?
The Autonomy Spectrum pattern replaces binary AI controls (on/off, assist/don't assist) with a graduated range of independence levels. Traditional AI interactions are either fully manual or fully automated, but agentic workflows demand nuance. A user might want their email agent to auto-sort messages without asking, but require explicit approval before sending any reply. This pattern provides four core levels - Observe & Suggest, Propose & Confirm, Act & Notify, and Full Autonomy - adjustable per task type. The key insight is that trust isn't global: users develop different comfort levels for different domains based on the agent's track record. By making autonomy granular and visible, this pattern prevents the all-or-nothing dynamic where a single bad experience causes users to abandon the agent entirely.
Example: Claude Code — Per-Tool Autonomy Control
Each tool the agent can use — file edits, git commands, test runners — can be set to Allow, Ask, or Deny independently. Developers grant full autonomy for low-risk actions while requiring confirmation for destructive operations.
AI Design Prompt
Want to learn more about this pattern?
Explore the full pattern with real-world examples, implementation guidelines, and code samples.
View Full Pattern