A discussion has emerged about enhancing user experience when AI responses are being generated in Discourse. BrianC raised the issue that there is no visual indication when an AI is processing a response, especially with slower models, leading to user confusion or frustration. He suggested a simple ‘Thinking’ or ‘Reasoning’ status, or leveraging the existing ‘user is typing’ indicator to show ‘AI is typing’. sam acknowledged the idea and noted that while it’s a complex feature, a straightforward implementation could use the current typing indicator. BrianC confirmed that the blue flashing dot, which typically indicates activity, disappears during AI thinking and only reappears when the response starts, which is not the expected behavior. He reported this issue occurs across multiple models, including Qwen3 30B A3B and Qwen3 235B, via OpenRouter. sam questioned the cause, indicating the behavior should not occur. The thread highlights a need for a clear visual cue to improve transparency during AI processing.
בדקתי את אפשרות ה-Triage עם Persona עבור תגובות אוטומטיות וראיתי שכאשר משתמשים במודלי חשיבה עם עיבוד רקע איטי יותר, אין סימן ויזואלי למשתמשים לדעת שמשהו בתגובה אוטומטית מתבצע. זה ישפר מאוד את חוויית המשתמש אם יוצג סימן תהליך—כמו “חושב” או “שיקול דעת”—בזמן שהמערכת מנסחת את התגובה.
האם קיימות תוכניות ליישם תכונה כזו? אני מאמין שהוספה כזו תעזור לנהל את הציפיות ותשפר את האינטראקציה הכוללת.
I moved this to feature, I don’t think this has popped up before, but it is certainly an interesting idea.
We have a lot of internals to support this kind of feature and this can happen as well outside of responders, eg @mentioning a persona that allows mentions in a topic.
Any additional consideration for this Thinking visual queue when AI thinking is causing delay? I think this is good for the Discourse presentation. Instead of a user being unaware of delay or frustrated they just chill and wait.
The current blue flashing dot indicator disappears when the AI starts to “think”. Is that the expected behavior? The indicator remaining visible and flashing would by itself be sufficient to show background activity occurring. It currently disappears during thinking and then reappears as the AI starts to respond on my Discourse instance.