Writing · Behavior map

How a proactive AI decides when to speak up

A proactive AI starts the conversation itself. Some decisions are made once, before it ships. Some are made again in every moment it is paying attention.

One-page map · October 2026

When the AI behaves badly, most teams add a line to the prompt. It is fast, and it often doesn’t hold, because many problems were never prompt problems. This map shows where each kind of decision lives, and where each kind of fix belongs.

One-page diagram titled How a proactive AI decides when to speak up. Top row: four steps decided before it ships (what it learns in training, the behavior plan, getting to know the person, rehearsal). Middle: a loop decided in every moment, where the system notices something, asks how sure it is and what it would cost to be wrong, then stays quiet, speaks up, asks, or acts, with one judgment driving both screen and voice. Below: three ways proactive goes wrong and the repair, a table of where fixes belong (training, behavior plan, code, getting to know the person, rehearsal), how to know it is working, and the need for one owner of the whole map.
Select the map to open it full size.

← Back to writing

Let’s talk model behavior

The interesting work is at the boundary.

If you’re building agentic or multimodal systems and the question of when a model should help, ask, or refuse is still unowned — that’s my seam.

© 2026 Alana Cortés — Model Behavior DesignerDesigned & built by Alana Cortés · Austin, TX