Pairing with AI in exploratory sessions

How it works

Pair testing works because a second person notices what the driver stops seeing, and suggests the next move. A model can take part of that role. During a session the tester describes observations as they go — screenshots and copied output help — and asks: what else could explain this, what input would test that idea, what would a malicious or confused user do here, have I tried this from another role? The conversation also leaves a running record of what was tried and seen, a head start on the session notes — though only for what the tester typed. It works best when the tester stays in charge of where the session goes and uses the model to widen options and to challenge premature conclusions.

When to use it

Solo exploratory sessions where there is nobody to pair with; when a session has gone stale and ideas have run out; when investigating a strange behaviour and needing alternative explanations; and when learning a new product, where the model's general knowledge covers some of the missing context.

Limitations

The model sees only what the tester describes, so its suggestions are filtered through the tester's own attention — it can't notice what the tester didn't mention. Constant prompting breaks the concentration that good exploration depends on, and switching between product and chat costs more than it seems. Suggestions are generic until the model has enough context, and it will confidently explain behaviour it has never seen.

What you still own

The direction of the session, what counts as a finding, and whether an explanation fits what you actually observed.

Cite this