AI onboarding teaches both the person and the AI
A first encounter between a staff member and an organisation's AI has two students; a useful onboarding step changes the user's understanding and writes to the AI's persistent context, and steps that achieve only one are filler.
When someone joins an organisation that uses AI, the encounter has two students, not one. The new staff member is learning the tool, and the tool is learning who the staff member is and what the organisation expects of it. The framing changes what counts as a useful onboarding step. A step that produces a change only in the user’s understanding is a video. A step that only configures the AI behind the scenes is a setup script. A useful step does both — and steps that achieve only one are filler regardless of how engaging they feel in the moment.
Why both outcomes are needed
The user-side outcome is the obvious one. The staff member leaves the encounter knowing how to use the tool, what is in scope, and what the organisation expects them to do with it. Without that, the rollout depends on memory and good intent, both of which decay (see Embed the AI policy in the AI itself).
The AI-side outcome is the less obvious one and the more leveraged. The same encounter is the opportunity to write the user’s role, function, and stable preferences into the AI’s persistent context, alongside the firm-wide rules — templates, regulatory or sectoral standards, internal vocabulary, communication style — that should apply to every later conversation. Without this step, the AI starts every later interaction effectively as a stranger, and the user pays the cost in repeated context-setting at the start of each conversation. The AI’s usefulness is bounded by the context it has been given (see Useful AI is a context problem); onboarding is the cheapest moment to load that context properly.
Design rules that follow
Three rules drop out of the dual frame.
Every step has both outputs. A welcome step that gathers no persistent identity is filler. A settings-configuration step that does not teach the user why is a setup script. The discipline is to specify, for each step, both outputs explicitly, and discard steps that have only one.
Hook before policy. Compliance content has the lowest engagement of any onboarding material. A live demonstration that the AI does useful work for the user’s actual function — running on the organisation’s templates, in the organisation’s standards — earns the attention that the policy section then asks for. Putting policy first codes the encounter as compliance training, and the AI gets read as compliance.
Identity-aware calibration. Self-reported seniority is not a clean proxy for tenure or expertise. A long-tenured specialist asked to identify as a “learner” or “junior” will be alienated before the encounter has done anything useful, regardless of how careful the rest of the design is. Map calibration to function and responsibility level, not to tenure-in-AI-tools. The analogous argument at programme-design scope is in Match AI programme ambition to working-team capability.
The dual-onboarding frame makes design decisions falsifiable. Each candidate step can be tested by asking what the user takes away and what the AI now persistently knows. A step that fails either test is not yet a step.