1 What is an AI agent?
Recent advances in large language models and retrieval-augmented generation have set the stage for a new wave of “agents.” An LLM agent is a system that uses an LLM as its reasoning engine to autonomously choose actions and pursue goals—the model acts as the agent’s brain. Thanks to LLMs’ broad generalization and emerging reasoning capabilities, agents can analyze context, form plans, and adapt to unfamiliar tasks rather than following rigid, predefined logic.
What makes an agent distinct is its tool use and dynamic decision-making loop: the LLM evaluates context, chooses a tool, observes results, and decides whether to continue or stop. Systems can exhibit different degrees of agency, from single LLM calls and fixed chains to routing, tool selection, multi-step termination control, and even tool creation. Because agentic workflows involve multiple calls and branching paths, they trade higher capability for cost, latency, and potential error compounding. The chapter offers practical criteria for deciding when to use an LLM (unstructured data, varied inputs) and when an agent is warranted (unpredictable steps, sufficient task value, manageable and detectable error risk).
The book takes a from-scratch, code-first path: begin with a basic ReAct agent (reason–act–observe) that selects and executes tools, then progressively add knowledge access (RAG) for long-term memory, planning for decomposing complex goals, and reflection for self-correction. It culminates in multi-agent systems that coordinate specialized roles, along with essential practices for monitoring and evaluation to debug, analyze costs and latency, and assess real-world performance. Readers are guided through this progression with Python-based implementations, supported by a minimal toolchain (local environment, API keys) and cost-aware development tips.
Example of a language model’s generalization capability.
User requests flow through the research agent, which branches into multiple searches and synthesis.
The LLM Agent's decision loop is an iterative process of LLM decision-making and tool use.
Progression of agency levels in LLM applications.
Progressive enhancement of agent architecture from basic LLM-tool integration (chapters 2-4) through advanced capabilities like memory, planning, and reflection (chapters 5-8) to multi-agent coordination (chapter 9).
Basic agent operational cycle demonstrating autonomous reasoning and tool use.
Agent with cognitive capabilities
Multi-agent roles for a software development task.
Summary
- An LLM agent is a program that determines the next action and evaluates goal completion (i.e., termination conditions) based on the output of an LLM, in accordance with a given objective and context.
- LLMs excel at generalization—handling diverse tasks without requiring task-specific retraining through capabilities such as few-shot learning, zero-shot learning, and reasoning with thinking tokens.
- Agency levels range from simple code execution to fully autonomous systems, with increasing autonomy as the LLM gains more control over task execution, action selection, and option determination.
- You should use an LLM agent for tasks that involve unstructured data analysis, require multiple unpredictable steps, have sufficient value to justify computational costs, and allow for error detection and correction.
- Building agents from scratch provides a deep understanding of core principles, clear visibility into the context flow, improved debugging capabilities, and the flexibility to create custom solutions that go beyond framework limitations.
Build an AI Agent (From Scratch) ebook for free
