Stardrift
an experience audit of Stardrift — AI travel search for people who fly a lot — reading the booking flow closely and proposing where it could be clearer.
what it is
Stardrift is AI travel search for frequent flyers — it plans and books the complicated itineraries, the ones that have to fit a real schedule and a real set of preferences.
Leila Clark brought me in early to audit the booking flow and find where it was losing people. what follows is that read — first the flow itself, then the slides, then what i took away from it.
general feedback




where the experience loses people — agent state, hierarchy, and formatting.
the booking flow







the booking flow, slide by slide, with redesign directions.
what i found
- the agent feels stalledyou can't tell when the agent is actually working — the text and icons sit static, so it reads as stuck. the status copy should update in real time and say what it's doing.
- the flight data is hard to trustnumbers show up without labels — which one is the seat count? fees change between screens, which quietly erodes trust. and a plain list makes comparison slow and mentally taxing.
- the hierarchy needs tighteningbooking buttons look like headers instead of actions, and the primary action isn't visually first.
- the formatting driftsseveral bullet styles, date formats that vary and feel off-standard, and reasoning text that looks too much like output and blends in.
- readability has roomtables would make flights easy to compare, and consistent formatting would lift both scannability and the sense of quality.
- the visuals mostly workthe line-and-dot motif is good, but it should be simpler and used more deliberately — and reasoning should be set apart from output, with block formatting, say.
a note of thanks
thank you to Leila Clark for the chance to work on Stardrift this early. it was a genuine learning experience, and i'm grateful to have spent time on something this ambitious.