What is an AI Agent?

An AI agent is a system where a language model decides which actions to take next, executes those actions using tools it has been given, observes the result and then decides what to do next. It runs in a loop until it either accomplishes its goal or hits a stopping condition.

That single sentence separates an agent from a chatbot.

The Four Ingredients

Every agent has four ingredients. Take any one of them away and it is not an agent.

A language model. The brain of the agent. It reads the current situation and decides what to do next. Claude, GPT and Gemini are the three most commonly used in production.

Tools. Functions the model can call to affect the outside world. Reading a file. Sending an email. Querying a database. Making an HTTP request. Without tools, the model can only produce text.

Memory. A record of what has happened so far in the current task. Usually this is the conversation history plus whatever the tools have returned. Without memory, each step is disconnected from the last and the agent cannot reason about what it has already done.

A loop. The mechanism that keeps handing the model the current state and asking "what next?" until the goal is met.

Agent vs Chatbot

A chatbot takes a message and returns a message. One turn in, one turn out. If the answer requires looking something up, either the chatbot cannot help or the developer has to build a separate step outside the chat.

An agent, given the same request, decides on its own to reach for a tool. It might query a database, discover the answer is not there, decide to search the web, still not find it, then respond with a clear "I could not find this" instead of hallucinating.

The difference is not the model. The difference is that the agent has been wrapped in a loop that lets the model take multiple actions before it responds to the user.

A Small Example

You ask an agent: "What is the balance on our staging AWS account?"

A chatbot would either guess or refuse.

An agent looks at its available tools, finds one called aws_billing_lookup, calls it with account="staging", reads the result, formats it and replies with the actual number.

The model did not need to know the answer at training time. It needed to know how to ask the right question of the right tool. That is what makes agents useful for anything involving live data, private data or actions with side effects.

Why This Matters Right Now

Agent frameworks were niche until late 2024. By mid 2026, 80 percent of Fortune 500 companies are running AI agents in production somewhere in their business. Job specs for AI developers now list agent orchestration, tool design and function calling as core requirements. The frameworks that support this (LangGraph, Claude Agent SDK, CrewAI and a handful of others) have gone from experiments to standard infrastructure.

Learning to build agents is not a specialisation any more. It is baseline AI developer literacy in 2026.

What This Course Will Cover

Over the next six modules you will:

1. Understand the common agent architectures and when each one fits
2. Build your first working agent with the Claude Agent SDK
3. Design tools an agent can actually use well
4. Move to LangGraph for complex workflows with branching, loops and persistence
5. Coordinate multiple agents that hand work off to each other
6. Build a capstone: a customer support agent that queries a ticket database and sends email

By the end you will have built and understood a working agent, and you will have code you can show as portfolio evidence that you have done real work with the frameworks employers hire for.

Key Takeaway

An AI agent is not a smart chatbot. It is a language model plus tools plus memory plus a loop. Once you have those four pieces working together, you can build systems that do genuine work in the world, not just produce clever text.