AI agent
An AI agent is a system that, beyond generating text, reads data and takes actions to finish a task. In support that means it checks the actual state of the customer account before answering, and when the answer involves doing something, it does it instead of explaining how.
What is AI agent?
An AI agent is a system that, beyond generating text, reads data and takes actions to finish a task. In support that means it checks the actual state of the customer account before answering, and when the answer involves doing something, it does it instead of explaining how.
Also: autonomous agent · AI support agent
The word "agent" means two things now
In support, "agent" used to mean the person answering. Since 2023 it also means the software that answers without anyone involved. Say "AI agent" for the second one: both live in the same team, and sentences like "the agent escalates to the agent" help nobody.
Why it matters
What changes in a SaaS
It is the difference between resolving and forwarding. A system that only generates text can explain anything that is written down; one that also reads and acts can close the case. In a SaaS, where half the questions depend on the plan, the permissions or the usage of one specific customer, that difference decides whether support volume drops at all.
AI agent in detail
The three pieces
A language model that writes, a way to retrieve information — your documentation and the account state — and a set of tools it can call. Remove the third and you have a chatbot; remove the second and you have a text generator that makes the data up.
What you decide when you set it up
What it may read and what it may do. That is a product conversation, not a configuration screen: "it can check the plan and the usage, it can resend an invoice, it cannot cancel a subscription." Skipping that decision is what produces agents that answer strange things.
Knowing when to stop
An agent that always answers is worse than one that recognises its limit. When the data does not add up, when judgement is needed or when the customer asks, the right move is to hand the conversation over with the case assembled.
How it is priced
This category usually charges per resolved conversation rather than per seat, because the system does not occupy a desk. It is an aligned model: if it does not resolve, it does not bill.
How Intake handles it
Questions about AI agent
What is the difference between an AI agent and a chatbot?
Where the answer comes from and what it can do with it. A chatbot answers from what is written in your documentation; an agent also reads the account state and can take actions on the systems you enable.
Does an AI agent make mistakes?
Yes. What separates a useful one from a dangerous one is how it fails: "I do not know, let me get a person" is acceptable; inventing a billing figure is not. Limiting where it can get data from is what prevents the second.
How long does it take to set one up for support?
Connecting the documentation takes hours. What takes time is deciding which endpoints it reaches and with what permissions — usually one or two afternoons of someone technical. If it is sold to you as a months-long project, what is being sold is the consulting.
Related terms
A term on its own is only half understood. These come up in the same conversation.
Chatbot
A chatbot is a program that holds a written conversation with a user. In support, current ones index the company documentation and answer from it; earlier ones followed decision trees written by hand.
Tool calling
Tool calling is the mechanism by which a language model can run functions you define instead of only writing. You describe the available tools, the model decides which one to use and with what arguments, and it receives the result to fold into its answer.
RAG — Retrieval-augmented generation
Retrieval-augmented generation, or RAG, is the technique of finding relevant information in your own sources and handing it to a language model so it writes the answer with that. The model does not memorise the content: it looks it up at answer time.
LLM — Language model
A language model is a system trained on enormous amounts of text that, given a fragment, predicts how it continues. Writing, summarising, translating and holding a conversation all come out of that one simple capability: they are all ways of continuing a text plausibly.
Deflection rate
Deflection rate is the percentage of questions resolved without a person getting involved, either because the customer found the answer themselves or because an automated agent closed it. It is the metric used to justify any investment in self-service or in AI.
Evals — Evaluation
An evaluation is a set of test cases used to check, repeatably, whether an AI system answers as it should. Each case carries an input and a success criterion, and the whole set is re-run every time something changes: the prompt, the model or the documentation.
