The chat advisor I designed on 30 July has one rule that mattered more than anything else in the build: not every question should cost the same.
Pulso’s suggestion engine calls Claude to research prices, and that costs real money, so before any paid lookup fires the app now shows a gate stating the cost and the roughly 45-second wait. Free questions, what’s my collection worth, what’s my shortlist look like, am I below target on anything, compute instantly in Swift from data already in memory. No model call, no wait, no charge.
Building the free path first surfaced a bug I’d have shipped otherwise: the monthly AI allowance was tracked globally, not per user, so one person’s questions could spend everyone’s budget for the month. Fixed the same day, scoped to user_id.
The advisor itself launched with a seeded opening turn and four price bands, same money, around, about half, much less, instead of a plain expensive or cheap toggle, each with a real reason attached, not just a number.
Still unsolved: the mode meant to fold this into one continuous chat thread, rather than a one-off lookup, hasn’t been wired up yet.
