Chatterfly

Chapter 26 of 36 · ~1 min

Copilot → Automation → Agent

AI adoption tends to move along a path. First a copilot: a person does the work and the model assists. Then partial automation: the model handles routine cases and a person handles exceptions. Then an agent: the model runs the process and a person supervises outcomes. Each step trades human control for speed and scale.

The path is not a ladder you must climb to the top. Some processes should stay at copilot for good. The right position depends on the cost of an error, how well the task is understood, and how much evidence you have about the model's behaviour on your own data. Moving one step at a time, on evidence, is how the trade-off stays honest.

Copilot, automation, agentCopilotperson works, model assistsAutomationmodel does routine, person handles exceptionsAgentmodel runs it, person superviseson evidenceon evidencemore human controlmore autonomy, speed, scaleNot a ladder to climb to the top. Some processes should stay at copilot for good.
Copilot, automation, agent: each step trades human control for speed and scale.

Exercise

Do it yourself

Place three of your processes on the path. For each, name the evidence that would justify moving it one step further.

Big question

Who decides when a process moves along the path, and on what evidence?