AI Agent

Autonomous Agent · LLM Agent

Definition

An AI agent is an LLM-based system that autonomously plans and executes multi-step tasks through tool and API calls, rather than stopping at a text response.

What makes an AI agent an agent

A language model on its own is an oracle: it answers questions but does not act. An AI agent is an
LLM plus three components:

Tools — access to external systems through function calling or MCP: search, APIs, databases,
browser interfaces
Memory — context preserved between steps: what has been done, what remains
A planning loop — the ability to decompose a goal into steps, execute them and revise the
plan from the results

The capacity to act across several steps and adapt is what separates an agent from an ordinary
model call.

Agent architecture

User goal
  ↓
LLM (planner + executor)
  ├── Reasoning (chain of thought)
  ├── Tools (function calling / MCP)
  │     ├── Catalogue search
  │     ├── Order API
  │     └── Payment token
  └── Memory (session context)
  ↓
Result

Chatbot versus AI agent

Property	Chatbot	AI agent
Output	Text	Action plus text
Autonomy	Reactive	Proactive
Tools	Usually none	Mandatory
Multi-step	No	The core of the concept
E-commerce use	FAQ, support	Discovery, purchase, personalization

Application in e-commerce

An AI agent changes the shopper’s role from active searcher to task setter. The person states what
they need; the agent performs the search, filtering, comparison and ordering through standard
protocols — MCP for data, commerce protocols for the order.

Important: the autonomy granted to an agent in financial operations has to match the risk. An
agent able to spend without limits is a serious security exposure. Payment mandates and network
agent tokens address this through explicit spending policies.

Frequently asked questions

What is the difference between an AI agent and a chatbot?

A chatbot answers questions from a script or generates text in response to a prompt. An AI agent acts: it plans a chain of steps, calls external tools such as APIs and databases, makes intermediate decisions and revises the plan when needed. An agent is proactive; a bot is reactive.

What does autonomy mean for an agent?

Autonomy is the ability to complete a multi-step task without confirming every step. The degree varies, from proposes a plan and waits for approval at the low end to executes the task entirely on its own at the high end. For financial operations in commerce, a human in the loop is the recommended design.

Which LLM sits underneath commercial agents?

Several. Different vendors use their own frontier models, and open-weight models appear in self-hosted solutions. The model choice affects planning and reasoning quality, but the interaction protocols do not depend on any particular LLM.