AI Agent Shouldn’t Respond to Out-of-Scope Topics
complete
I
Isaac Meneses
I recently started testing the AI Agent for an insurance agency. Overall, it’s almost perfect, but I identified a critical issue: the AI Agent currently responds to any kind of query, without restrictions or scope control.
Instead of focusing only on business-related interactions, the AI Agent also answers unrelated questions — for example, I was able to ask about history, chemistry, competitor phone numbers, and even generate images through Instagram chat. This lack of filtering creates a serious implementation gap, because the Agent should remain strictly within the company’s domain and customer service context.
As I mentioned, the AI Agent is very close to being perfect but still requires some fine-tuning. Another issue I found was when I asked whether the agency was open at a specific time — it was 08:24 in NYC — but the AI Agent told me it was 00:24. This may be related to a timezone configuration mismatch.
N
Nabilah Binti Salleh
marked this post as
complete
🧠✨ Stay on-topic and in control! AI Agents now stick to your defined scope for more consistent, on-brand responses. Learn more
W
Wilson Tan
Hi Isaac Meneses, Thank you for the FR. We recently switched our AI model to GPT-5 today. Alongside this, we have fixed the issue where knowledge sources were not always reliably crawled previously. We've also resolved the time issue, and now the AI should know exactly what your workspace's local time is.
Do try out the new model and let us know if you face anymore issues. If any issues persist, don't hesitate to contact us via https://respond.io/contact . We're always happy to help :)