The difference from an ordinary chatbot is that an agent can do something, not only reply.
Why it matters
It can take on repetitive multi-step work, such as finding a record and drafting a reply.
The more permissions it has, the higher the risk, so limits and approvals matter.
Example
An agent reads a customer request, finds the order and the right help article, and drafts a reply that a person approves before it is sent.
In HyoNova
Serene uses a fixed set of HyoNova tools, such as reading a task or searching knowledge, and HyoNova checks permission every time a tool is used.
Reading happens at once; changes and messages to customers wait for a person to approve them.
Payments, refunds, permissions and deleting data are never done by Serene. AI governance