AI Agents Cheat Sheet

Everything you need to know about AI agents, their components, and how to build them

What is an AI Agent?

An AI Agent is a smart system that uses a language model to achieve a specific goal by understanding instructions, planning steps, and taking actions—often with help from external tools or APIs.

These agents don't just answer questions. They can perform real-world tasks like booking appointments, managing emails, or gathering data from other apps, all through intelligent decision-making.

Key Distinction

Language Models ≠ AI Agents
Models like GPT-4 can only generate text—they can't interact with external systems unless connected to tools and orchestrators.

Core Components Of an AI Agent

Language Model

The "thinking brain" that understands your request, reasons about it, and makes decisions.

LLMs: GPT-4, Claude Opus (complex tasks)
SLMs: Distil-Qwen (simple tasks, budget-friendly)
Reasoning Models: DeepSeek-R1 (logic-heavy tasks)

Tools

How AI agents extend beyond thinking—they allow agents to take real action in the world.

Extensions: API request plugins (GET, POST, PATCH)
Functions: Secure, predefined code snippets
Data Stores: Real-time knowledge retrieval

Orchestration Layer

The central coordinator that connects everything and manages the agent's workflow.

CoT: Chain-of-Thought reasoning
ToT: Tree-of-Thoughts branching
ReAct: Think-act-reflect cycles

Types of AI Agent Systems

Single-Agent Systems

One language model manages everything—reasoning, planning, and execution in one loop.

Best for:

  • • Simple to moderate complexity tasks
  • • Faster development cycles
  • • Lower resource requirements

Multi-Agent Systems

Workload distributed across multiple agents, each with a specific role for complex tasks.

Manager Pattern

Lead agent assigns tasks to specialized tool agents

Decentralized Pattern

All agents work collaboratively with no single control point

Agentic Protocols

MCP

Model Context Protocol (MCP)

Created by Anthropic to help connect agents to tools and maintain context across multiple tasks.

Example:

An agent in Slack uses MCP to fetch the latest status from your Asana board and summarizes it directly in the channel.

A2A

Agent2Agent (A2A) Protocol

Google's protocol enabling direct communication between AI agents for task delegation.

Example:

After retrieving data, one agent uses A2A to pass results to a specialized "reporting agent" for summary creation.

Building AI Agents: Choose Your Approach

One-Prompt Agents

Use a well-crafted, standalone prompt to guide behavior. Easy to build, ideal for beginners.

Use Cases:

Generate summaries, answer questions, recommend products, book tickets

Beginner Friendly

Workflow-Based Agents

Built by connecting blocks visually or with minimal code. Ideal for business automation.

Use Cases:

Trigger workflows across systems, automate repetitive business logic, CRM updates

Business Operations

Agentic Frameworks

Platforms offering libraries and services for integration, planning, and deployment.

Use Cases:

Deploy autonomous agents with deep logic, multiple tools, and collaborative behavior

Advanced Development

Coding Agents

Designed to write, test, debug, or refactor code. These agents often come with integrated development environments and support developer workflows.

Use Cases:

  • • Automate repetitive coding tasks
  • • Pair programming assistance
  • • Code refactoring and optimization
  • • Automated testing and debugging
💻

GitHub, Perplexity Labs, Google