AI Experience Makes Agents Legible

The next agent breakthrough may be better interaction design, not more autonomy. Powerful systems remain difficult to adopt when users cannot understand their intent, monitor their actions, or recover from mistakes.

• –Legibility should include clear goals, plans, tool calls, state changes, and failure explanations
• –Trust requires calibrated autonomy, with approvals and human handoffs for consequential actions
• –Developers need observability that serves both debugging and end-user understanding
• –Persistent memory should be inspectable, editable, and bounded by explicit permissions
• –AI Experience could become a useful counterpart to developer experience as agents become first-class collaborators