artificial intelligence and intelligent agents: What They Are and How They Work Together

If you have spent any time around AI lately, you have heard the term "AI agent" thrown around a lot. Coding agents, research agents, customer support agents, personal assistants. They are everywhere, and they all trace back to two foundational ideas: artificial intelligence, and the intelligent agent. Here is what each one means and how they fit together.

What Is Artificial Intelligence, Briefly

Artificial intelligence is the broad field of computer science concerned with building machines that can perform tasks which normally require human intelligence. That includes learning from data, reasoning about problems, understanding language, recognizing images, and making decisions under uncertainty.

Most AI in the real world is narrow AI: systems that are very good at one kind of task. A spam filter, a translation model, a recommendation engine, and a chess program are all narrow AI.

The important thing for this article is that artificial intelligence is the whole umbrella. Intelligent agents are one of the most important ideas underneath it.

What Is an Intelligent Agent

In artificial intelligence, an agent is anything that can be viewed as perceiving its environment and acting upon it to achieve its goals. The environment gives the agent inputs (percepts) through sensors, the agent decides what to do, and it carries out actions through actuators.

That definition is deliberately wide. A thermostat is an agent: it perceives temperature and acts by switching the heater on or off. A chess program perceives the board state and acts by choosing a move. A self-driving car perceives the road through cameras and lidar and acts by steering and braking.

The textbook word for a good agent is "rational": a rational agent tries to act in a way that achieves the best outcome for its goals, given what it knows. Intelligence, in this framing, is the ability to perceive, decide, and act toward a goal, and to do that well.

The 5 Classic Types of Intelligent Agents

AI textbooks (Russell and Norvig's is the standard one) describe five types of intelligent agents, each more capable than the last. You will see these on exams, in interviews, and in documentation for agent frameworks.

1. Simple reflex agents. These act on the current percept alone, using condition-action rules ("if X is true, do Y"), with no memory of anything that happened before.

They are fast and simple, and they work fine in fully observable environments, but they fall apart the moment they need to remember the past.

2. Model-based reflex agents. These keep an internal model of how the world works, so they can track parts of the environment they cannot see right now and update their picture of the world as new percepts arrive.

A robot vacuum that maps a room as it moves is a model-based reflex agent: it cannot see the whole room at once, so it maintains a model of where it has been and where obstacles are.

3. Goal-based agents. These do more than follow rules. They have a description of a desired goal state, and they choose actions that move the environment toward that goal, often by searching through possible sequences of actions.

A GPS navigator is goal-based: it knows your destination, considers many possible routes, and picks the sequence of turns that gets you there. Simple reflex rules alone could never plan a route.

4. Utility-based agents. These go one step further and attach a utility function (a measure of desirability) to possible outcomes, so they can weigh tradeoffs and choose the action that maximizes expected utility, not just any action that reaches a goal.

When your navigation app picks a slightly slower route that avoids a toll road, that is utility-based thinking: it is trading off time, cost, and comfort instead of blindly optimizing one thing.

5. Learning agents. These start with incomplete knowledge and improve over time by learning from experience. They are usually described as having four parts: a learning element that improves the agent's knowledge, a performance element that picks actions, a critic that judges how well things went, and a problem generator that suggests new things to try.

Every recommendation system that gets better the more you use it, and every game-playing AI that improves through practice, is a learning agent.

In practice, real systems mix these types. A modern assistant might use goal-based planning for a task, utility-based tradeoffs when picking between options, and learning from your feedback over time.

The PEAS Framework

When AI engineers design an agent, they describe it with the PEAS framework: Performance measure, Environment, Actuators, Sensors.

• Performance measure: how do we judge success? (For a vacuum agent: how much dirt it picks up. For a coding assistant: working code that passes tests.)
• Environment: what world does it operate in? (A house, a codebase, the web.)
• Actuators: what can it do? (Move, suck dirt, write files, run commands, send messages.)
• Sensors: what can it perceive? (Cameras, temperature readings, file contents, terminal output, your chat messages.)

PEAS is useful because it forces you to be concrete. "Build me an AI agent" is vague. "Build an agent whose environment is our support inbox, whose sensors read incoming tickets, whose actuators draft and send replies, and whose performance is measured by resolution rate" is a specification you can actually build.

Today's LLM Agents Are Intelligent Agents (With One Big Weakness)

Here is where the textbook meets your daily life. The coding agents and AI assistants everyone uses now, tools like Claude Code, Codex, Cursor, and the chat assistants on your phone, are intelligent agents in exactly the classic sense. They perceive their environment (your messages, your files, terminal output, browser pages), they act on it (writing code, running commands, editing documents, calling tools), and they work toward goals (fix this bug, build this feature, answer this question).

They also share one big practical weakness, and it is the same weakness every one of these tools has: they lose context between sessions and between tools. Close the chat, and the next session starts cold. Use two different tools on the same project, and each one has its own separate memory of what happened. Your coding rules, your project decisions, the plan you worked out yesterday, the reason you chose one approach over another, all of it gets trapped inside whichever tool you happened to be using last. The agent loop of perceive, decide, and act works fine inside a single session, but the memory of what was perceived and decided does not travel with you.

This is the problem a shared memory layer solves. Vilix AI is cloud-hosted, so there is zero infrastructure to manage, and over MCP it gives the same memory and context across your phone, your laptop, and every connected AI tool. It stores full conversation history, not just extracted facts, so you can revisit the real discussion whenever you need it. Your data is portable, export everything or delete it anytime, there is a free plan forever, and the seven-day Pro trial needs no credit card. One honest caveat: MCP memory tools only run when the model decides to call them, so occasionally you need to tell the agent "check Vilix AI for context first." With that small habit, the agent stops forgetting and starts building on what it already knows about your work.

FAQ

What is an intelligent agent in artificial intelligence?

An intelligent agent is anything that perceives its environment through sensors and acts on it through actuators to achieve its goals. The key loop is perceive, decide, act, repeat. Examples range from thermostats to self-driving cars to modern AI chatbots.

What is the difference between artificial intelligence and intelligent agents?

Artificial intelligence is the entire field of building machines that perform tasks requiring intelligence. An intelligent agent is a specific concept within that field: a system structured around perceiving an environment and acting toward goals. You can think of AI as the discipline and intelligent agents as one of its most important design patterns.

What are the 5 types of intelligent agents?

The five classic types are simple reflex agents (act on current percepts with condition-action rules), model-based reflex agents (maintain an internal model of the world), goal-based agents (plan actions toward a desired goal), utility-based agents (maximize a measure of desirability across tradeoffs), and learning agents (improve from experience over time).

What is PEAS in artificial intelligence?

PEAS is a framework for specifying an intelligent agent: Performance measure (how success is judged), Environment (the world it operates in), Actuators (what it can do), and Sensors (what it can perceive). It turns a vague "build an AI agent" request into a concrete design.

Are chatbots intelligent agents?

Yes. A chatbot perceives your messages, decides on a response, and acts by producing text, all in service of the goal of helping you. Modern chatbots and coding assistants are some of the most widely used intelligent agents in existence.

Why do AI agents forget context between sessions?

Most AI agents keep their memory inside the session or the tool itself. When the session ends or you switch tools, that context does not carry over, because there is no shared, persistent memory layer connecting them. Each session starts cold and each tool keeps its own silo, which is why a shared memory layer that travels across tools and devices makes such a difference.

The Takeaway

Artificial intelligence is the field, and the intelligent agent is one of its central ideas: perceive the environment, decide, act toward a goal, repeat. The five classic agent types and the PEAS framework give you the vocabulary to understand and design agents, and today's LLM tools are those same ideas running at a scale the textbooks never imagined. Their remaining weakness is not intelligence, it is memory. Give your agents a shared memory that follows you across every tool, and the loop finally works the way it was always supposed to.

If you work with more than one AI tool, Vilix AI keeps their shared context in one place so nothing gets lost between sessions: https://vilix.ai