Every AI response has a price. Users discover it when a limit hits or a bill arrives. This UX Collective guide by frames AI cost design around two variables: the unit users see themselves spending, tokens, dollars, credits, tasks, or outcomes, and the moment they receive that information, before, during, or after the work. The core argument is blunt: showing raw token counts fails users. '382,000 tokens used' tells no one whether their next research run fits within their allowance or what to change.
The technical foundation matters here and the article earns its length explaining it. Output tokens cost roughly five times more than input tokens on several current models. A 15-word prompt runs about 20 text tokens. A single 1092x1092 image runs roughly 1,600. Five minutes of video can consume 30,000 to 90,000 tokens depending on resolution. These cannot be priced as equivalent units, yet a single input box treats them identically. The piece also cites Lambrecht and Skiera (2006), a study of 10,882 internet subscribers, to explain why flat-rate predictability beats metered precision: people enjoy a service less when they watch the meter climb.
The unit choice determines who carries uncertainty. Tokens and metered dollars push that uncertainty onto users. Task and outcome pricing shifts it to vendors. Usage allowances remove per-action math but obscure the limit. The article is worth reading in full for its breakdown of hybrid pricing models and the specific design patterns for surfacing cost before, during, and after an agentic workflow, a problem that only gets harder as AI products replace single-button actions with multi-step autonomous runs.
[READ ORIGINAL →]