What are the limitations?
Your agent is only as good as the knowledge you give it, it will say when it doesn't know rather than invent an answer, and anything sensitive should still pass a person.
Last reviewed 29 September 2026
Where we'd set expectations
The agent is strong at the questions you answer every day and weaker at anything genuinely novel — one-off commercial terms, complaints, or anything that needs judgement. It won't invent an answer to cover the gap; it tells the customer it will check and passes the conversation on.
- It needs a person's approval for outbound messages until you turn that off
- Complex or regulated advice should stay with a human — we'll tell you which parts of your process should not be automated
- It answers in the languages covered by your knowledge, so multilingual coverage follows your sources
- If your documents conflict, its answers will be inconsistent — one authoritative source per subject
How we help you avoid the gaps
We review your first conversations with you and tell you honestly which parts of your process are a good fit today and which aren't. We'd rather scope an agent tightly and have it be reliable than promise it can do everything.
Did this answer your question?
Your feedback tells us which articles to improve next.
Related articles
Still stuck? milly.patterson@agenttrac.ai — we reply within one working day.
/help/what-are-the-limitations