Portraits can produce voice without producing dependable behavior.

We tested whether a detailed identity portrait was enough to make an AI agent follow a specific operating practice. The portrait made the agent sound more recognizable, but it did not reliably produce the behavior being tested. Explicit operating instructions improved the result.

The experiment changed how we build. Identity can shape how an agent sounds; dependable practice also needs procedures that can be observed, corrected, and tested. We are continuing to test the boundary between recognizable identity and dependable practice.