Most organizations treat AI autonomy as a configuration decision: which tools get which permissions, set by IT and forgotten. However, the variable that determines whether a team can safely hand work to an agent is a person's demonstrated judgment: how they brief the AI, whether they catch its wrong answers, when they trust versus verify, what they keep and what they delegate. That is a competence — L&D's jurisdiction.

This session lays out a six-profile progression (i.e. Observer, Operator, Collaborator, Delegator, Orchestrator, Steward) in which each rung is a richer working model of the AI: from treating it as a black box whose inputs map to outputs, to predicting what it will do with novel input, to knowing where it breaks, to composing it into governed pipelines, to intervening in agent runs neither by rubber-stamping nor micromanaging. It shows what "ready for more autonomy" observably looks like at each rung and how to evaluate it from people's real work with AI rather than quizzes or self-report. It draws on building Evolve Coach at Nebius Academy, which coaches engineers inside the coding assistants they already use.

During this session, you will learn:

• The six-profile progression
• The observable behaviors behind each profile
• How to use the template for putting autonomy on L&D’s roadmap