AI agents, chatbots and fixed rules: the difference
An agentic system runs a whole part of a business, end to end. AI agents share the work, from the first enquiry to the final invoice, on top of the company's own records. People decide the cases that need a person and approve anything that leaves the company.
It is not a chatbot
A chatbot answers questions. An agentic system does the work behind the answer: it prepares the quote, follows up on the enquiry and keeps the records up to date. A person approves anything that leaves the company.
It is not fixed-rule automation
Fixed rules break when something changes: a new customer, a new product, a message that does not fit the form. Agents read what comes in and choose the next step, inside limits the company sets.
Where people decide
By default, an agent prepares the work and waits, and a named person approves it before anything leaves the company or any money moves. An agent can act alone only inside narrow limits set in advance, such as known customers and an amount cap.
Where the term comes from
We use the term the way Anthropic does in its guide to building agents. "Agentic systems" covers both workflows, where code sets the path, and agents, where the model directs its own steps [1].
Sources
- Anthropic, "Building effective agents", 19 Dec 2024. anthropic.com/research/building-effective-agents