Phase 07 · AI agents Core

AI Agents Explained

An agent is an LLM wrapped in a loop that can choose actions and use tools.

In plain terms

A chatbot replies once. An agent can decide to search, read a result, call a calculator, update its plan, and continue until it has enough evidence to answer.

Why it matters

Some tasks cannot be solved in one model call because the next step depends on the previous result.

How it works

The loop is simple: send the task and available tools, let the model choose a tool, run that tool in code, append the result, then ask the model what to do next.

When you use it

Use agents for open-ended workflows such as research, coding, data cleanup, browser tasks, and multi-step operations.

Common mistakes

  • Calling every tool workflow an agent even when a normal function call is enough.
  • Letting agents take high-impact actions without confirmation.
  • Skipping traces, which makes failures impossible to debug.

Best practices

  • Start with a small tool list.
  • Log every thought-free action, tool call, and result.
  • Put hard limits on steps, time, and spend.

Try it yourself

Build a research helper with search and note-taking tools. Limit it to five steps and print the trace.

Deep dive

An AI agent is an LLM inside a loop that can choose actions, use tools, observe results, and continue until it reaches a goal.

What it is

A normal chatbot answers once. An agent can search, calculate, retrieve documents, call APIs, update state, and decide what to do next.

Why it matters

Agents are useful when a task cannot be solved in one model call because each next step depends on previous results.

How it works

The app sends the task and tool definitions to the model. The model requests a structured tool call. Your code validates and runs the tool, sends the result back, and the loop continues.

Common mistakes

  • Giving agents too many tools.
  • Allowing high-impact actions without confirmation.
  • Skipping traces and step limits.

Best practices

  • Start with one or two tools.
  • Log every action and observation.
  • Set hard limits for steps, cost, and time.

Practical workflow

goal -> choose tool -> run code -> observe result -> decide next step -> final answer

Resources

FAQ

Are AI agents autonomous?

They can be partially autonomous, but production agents should have limits, validation, and human confirmation for risky actions.

Do agents always need memory?

No. Add memory only when the task needs continuity across steps or sessions.

Related topics

Tool use / function calling Letting the model request actions - search, calculate, query a database - by emitting structured calls to functions you define and execute. Structured output (JSON mode) Forcing the model to answer in an exact machine-readable shape - usually JSON matching a schema - so code can consume it. Prompt injection and AI security Prompt injection is malicious instruction text hidden inside data the model reads. Prompt caching Providers cache the unchanged prefix of your prompt so repeated calls reprocess it at ~10% of the price and much lower latency. Chain of thought & reasoning models Getting the model to work through a problem step-by-step before answering - either by prompting it to, or by using models trained to reason internally. Tools and function calling Tool calling lets a model request structured actions that your code executes. Agent executor, ReAct and toolkits An Agent Executor runs an Agent loop: Reasoning chooses an Action, a Tool runs, an Observation returns, and Decision Making selects the next step. Agentic AI goals, tasks and context Agentic AI is AI that works toward a Goal by breaking it into Tasks, choosing Actions, reading Observations and updating State inside an Environment. Capstone production AI product One deployed project that combines retrieval, tools, evals, security, cost tracking, and documentation.