prompttested
Show the work before the agent finishes the sentence
Small progress events can make a slow AI workflow feel understandable without inventing fake certainty.
An agent may need to search, call tools, and combine results before it can write the final answer. A spinner hides all of that and leaves the person guessing whether the system is thinking, stuck, or quietly on holiday.
A better experience could stream honest milestones: searching, checking a source, calculating, drafting. These are not theatrical chain-of-thought messages; they are simple product events the system already knows. The result is still allowed to take time, but the waiting becomes legible.
Context: Reusable prompt from this candidate
Review this AI workflow and propose a small set of user-facing progress events. Each event must correspond to a real application state or tool action, avoid exposing private reasoning or data, and use plain language. Return an event name, when it fires, the user-facing message, and how failure is shown. Do not implement anything until I approve the event map.