When people talk about “using AI”, they often speak as if it were a single thing with a single role.
As if there were one correct way to relate to it.
In practice, AI slips into very different roles — sometimes without us noticing. And many of the problems we run into don’t come from the tool itself, but from a mismatch between the role we expect it to play and the role it is actually playing.
So before asking what AI can do, it may help to ask a simpler question:
What kind of partner is this — right now?
Sometimes AI behaves like a teacher.
It explains, rephrases, gives examples, fills gaps in understanding. Used well, this can be empowering. Used carelessly, it can create the illusion of learning without the effort that makes learning stick.
Sometimes it plays the role of a researcher or librarian.
It searches, summarises, compares, retrieves. This is enormously useful — as long as we remember that gathering information is not the same as judging it, and synthesis is not the same as insight.
At other times, AI becomes a historian, placing things in context, tracing long arcs, reminding us that very little is truly new. This role can be grounding — unless we mistake narrative coherence for inevitability.
There are also more interpretive roles.
AI can act like a psychologist, naming patterns and dynamics.
Or like a tarot reader or astrologer, offering symbolic mirrors rather than answers — prompts for intuition, reflection, timing.
These roles don’t tell us what to do. They help us see ourselves thinking.
Then there are the creative roles: artist, writer’s companion, idea generator. Here AI can provoke, remix, surprise. The risk is not error, but sameness — letting its voice quietly replace our own.
And sometimes — often by default — AI is treated as a problem-solver.
Give it a task. Get a solution. Follow the steps.
This is where misunderstandings multiply.
Problem-solving assumes stable context, persistent memory, and shared understanding of the goal. When those assumptions fail, the help evaporates — or worse, gives false confidence. Instructions without presence are not partnership.
There are also roles best approached with caution:
the devil’s advocate, the mad scientist, the system that asks “what if we push this further?” Useful — but only when consciously invited, never when left on autopilot.
The trouble begins when one role silently takes over all the others.
When the teacher becomes the authority.
When the assistant becomes the decision-maker.
When the partner becomes an autopilot we never turn off.
Some tools are most useful when they are always available.
Others are most useful when they are summoned deliberately — and closed again when the work resumes.
Choosing how to work with AI is not about finding the right setting once and for all. It’s about staying aware of the role it is playing — and whether that role still makes sense for the moment we’re in.
The question is not “Can AI do this?”
It’s “Is this the kind of help I want right now?”
That question, asked often enough, may matter more than any answer AI can give.