- Chatbot
- Answers from approved knowledge
- Agent
- Takes actions in your systems
- Cost driver
- Actions, not answers
If the work is answering, you need a chatbot
When the volume is questions with existing answers (opening hours, policy, specification, delivery), the right build is a well-grounded chatbot reading your approved content. It is faster to deploy, cheaper to run and easier to govern.
The quality bar is grounding: answers must come from your documents and data, not from the model’s general knowledge. That is the difference between a helpful assistant and a confident liar.
If the work is doing, you need an agent
The moment the system has to check a diary, create a booking, write to a CRM or send something outward, it is an agent. It now makes decisions with consequences, which brings approval points, thresholds and an audit record.
Answers are cheap to govern. Actions are not.
Most real systems are both
In practice the useful build answers the common questions and acts on a narrow set of well-understood cases, escalating everything else. Starting narrow on actions is what keeps the first eight weeks calm.
Nexvix scopes the answer set and the action set as two separate lists, because they carry different risk and different cost.
Four questions that settle it
- Does the work end in an answer, or in a change to a record?
- Would a wrong instance cost a follow-up, or money?
- Does it need data that lives behind an API?
- Who approves the consequential case, by name?
Two or more “change to a record” answers and you are buying an agent. Price accordingly.