Spinners are broken for AI. For thirty years, interface designers used progress indicators to signal one thing: data is moving across a network. AI agents introduce a categorically different wait. When a scheduling agent pauses for twenty seconds, it is evaluating options, cross-referencing calendars, and constructing a plan. Showing a spinner during that process tells users nothing. It creates anxiety, not confidence. The fix is not visual. It is linguistic.
The agentic update formula has three required parts: an action word, a specific item, and the active constraints. 'Searching for flights' fails all three. 'Scanning Lufthansa and United for anything under $600' passes all three. Perplexity AI already does this correctly, surfacing a live list of completed sub-steps as the model works. The formula scales: a scheduling agent should display at minimum four discrete steps, from checking the requester's calendar, to cross-referencing attendees, to confirming the booked slot and triggering the email invite. Each step names who and what is involved. The user never has to guess whether the agent remembered the context of the request.
Tone is a risk variable, not a brand choice. The Impact and Risk Matrix from Part 1 of this series determines whether the interface uses conversational language or mechanical precision. A calendar assistant can sound human. An interface managing a financial transfer that says it is 'thinking hard about your money' will cause panic. The matrix is a starting point, not a rulebook. A/B tests, usability studies, and direct user interviews are required to validate which words build trust and which create stress for your specific user population. The full article covers how to apply this framework across the Decision Node Audit output and is worth reading for the scheduling and travel booking worked examples alone.
[READ ORIGINAL →]