Short answer
A chatbot generates replies. An agent takes actions in your systems — reading a record, checking rules, updating state, and escalating what it cannot resolve. The practical difference is write access and a defined scope of authority. If a system cannot change anything, its ceiling is deflection, no matter how well it writes.
Why companies feel let down by chatbots
A common experience: a business deploys a well-built assistant on its website, the conversations read fluently, and support volume barely moves. The usual conclusion is that the AI was not good enough. Usually it was fine — it simply had nothing to act on.
Consider what customers actually contact you about. Where is my order. Can I move my appointment. Has my invoice been received. Can I change the delivery address. Every one of those requires reading live system state, and most require writing to it. A system with no integration can only describe the policy for changing an address; it cannot change the address.
What makes something an agent
- It can read live state from the systems that hold the truth — CRM, ERP, ticketing, calendar, billing.
- It can write to them, within limits that are written down rather than implied.
- It knows what it is not allowed to do, and escalates instead of guessing.
- It hands off with context — the thread, the record, the reasoning, and a proposed next step.
- It leaves an audit trail showing what it did and on what basis.
Scope of authority is the real deliverable
Before an agent is built, someone in the business should be able to answer, precisely: what may it do unattended, what must it propose for approval, and what must it never attempt. That document matters more than the model choice. It is what makes the system defensible when a regulator, an insurer or a customer asks how it works.
Design the blast radius before you design the capability.
How to tell what you are being sold
Three questions cut through most vendor conversations. Which of our systems will it write to? What happens when it is not confident? Can you show me the log of what it did and why? A vendor selling a chatbot will answer the first question with training data rather than integrations.
Chatbot
- Question
- Generates a reply
From training, not your data
- Customer still waiting
Agent
- Question
- Reads the record
- Checks the rules
- Acts, or proposes
Within written authority
- Escalates with context
| Ask the vendor | A chatbot answer sounds like |
|---|---|
| Which of our systems will it write to? | Talks about training data instead |
| What happens when it is not confident? | Reassures you about accuracy |
| Can you show the log of what it did and why? | Describes the conversation history |
A chatbot can describe your refund policy. It cannot issue the refund. Since most contact requires reading and writing live records, that ceiling decides the outcome.
Last updated 2026-08-20 · Koderead
Related
Apply this to your own business
Eleven questions, scored in your browser, with the arithmetic shown.
Find my AI opportunities