Skip to main content

🤖 The Fundamentals of Agentic Systems

· 24 min read
Omar Eid
Omar Eid
Software Engineer
Cursor
Cursor
GenAI Agent

I wanted to document the fundamentals of Gen AI that I have been iteratively learning about; thus I made this post with the help of gen AI :)

This comprehensive glossary explains essential GenAI concepts, why they matter, and provides practical implementation examples. It focuses on clarifying the differences between commonly confused components.

What This Guide Covers:

  • Essential GenAI concepts and their importance
  • Practical implementation examples for each concept
  • Key differences between similar-sounding components
  • Interactive learning diagram with clickable navigation
  • Real-world applications and use cases

GenAI Domain Overview​

GenAI Fundamentals Glossary​

Foundation Models (LLMs)​

What it is: Large language models that can understand and generate human-like text.

Why it matters: These are the "brains" that power your AI system. Without them, you can't have AI.

Key Characteristics:

  • Trained on massive datasets (trillions of tokens)
  • Capable of text generation, reasoning, and conversation
  • Foundation for all other GenAI components

Prompting​

What it is: The instructions you give to AI models to get them to do what you want.

Why it matters: Good prompts = good results. Bad prompts = useless results.

Key Characteristics:

  • Clear, specific instructions
  • Context and examples when needed
  • Iterative refinement process
  • Foundation for all AI interactions

RAG (Retrieval Augmented Generation)​

What it is: A way to let AI access your company's documents, databases, and knowledge to give better answers.

Why it matters: Without RAG, AI only knows what it learned during training. With RAG, it can access your specific information.

Key Characteristics:

  • Combines information retrieval with text generation
  • Provides access to real-time, specific data
  • Reduces hallucinations by grounding responses in facts
  • Essential for enterprise AI applications

Vector Databases​

What it is: Special databases that can find similar information even when you don't use the exact same words.

Why it matters: Humans search by meaning, not exact words. Vector databases let AI do the same.

Key Characteristics:

  • Store high-dimensional vectors (embeddings)
  • Enable semantic search and similarity matching
  • Optimized for approximate nearest neighbor (ANN) search
  • Essential for RAG and semantic search applications

Memory Systems​

What it is: Ways for AI to remember what happened in previous conversations.

Why it matters: Without memory, every conversation starts from scratch. With memory, AI can build on past interactions.

Key Characteristics:

  • Persistent storage across conversations
  • Context retrieval and injection
  • User preference learning
  • Essential for personalized AI experiences

Agentic Systems​

What it is: AI systems that can not just talk, but actually do things like send emails, update databases, or make API calls.

Why it matters: Most AI just talks. Agentic AI actually gets things done.

Key Characteristics:

  • Autonomous task execution
  • Tool and system integration
  • Multi-step reasoning and planning
  • Real-world action capabilities

Tool Integration​

What it is: Ways to connect AI to your existing tools and systems (databases, APIs, software).

Why it matters: AI is only useful if it can work with your existing systems and data.

Key Characteristics:

  • Standardized interfaces for tool access
  • Secure authentication and authorization
  • Real-time data access and manipulation
  • Essential for practical AI applications

Evaluation and Testing​

What it is: Ways to test and measure whether your AI system is working correctly.

Why it matters: AI can seem to work but actually be wrong or harmful. Testing helps catch problems.

Key Characteristics:

  • Systematic evaluation of AI performance
  • Automated testing and validation
  • Human feedback integration
  • Continuous improvement through metrics

Security and Guardrails​

What it is: Systems that prevent AI from doing harmful things or accessing sensitive information.

Why it matters: AI can be powerful but also dangerous if not properly controlled.

Key Characteristics:

  • Content filtering and safety controls
  • Access control and authentication
  • Data protection and privacy
  • Compliance and audit capabilities

MCP (Model Context Protocol)​

What it is: A standard way for AI systems to connect to tools and data sources.

Why it matters: Without standards, every AI system needs custom connections. MCP makes it easier to connect AI to different tools.

Key Characteristics:

  • Standardized protocol for tool access
  • Interoperability between different AI systems
  • Secure and controlled tool access
  • Extensible architecture for new tools

Infrastructure as Code (IaC)​

What it is: A way to manage and provision infrastructure using code instead of manual processes.

Why it matters: IaC makes GenAI deployments reproducible, scalable, and maintainable by treating infrastructure as software.

Key Characteristics:

  • Version-controlled infrastructure definitions
  • Automated deployment and configuration
  • Consistent environments across development, staging, and production
  • Easy rollback and disaster recovery

How These Concepts Work Together​

Think of building an AI system like building a house:

  • Foundation Models = The foundation (everything else builds on this)
  • Prompting = The blueprints (how you tell it what to do)
  • RAG = The library (access to information)
  • Vector Databases = The filing system (organizing information)
  • Memory = The family history (remembering past interactions)
  • Agentic Systems = The hands (actually doing things)
  • Tool Integration = The utilities (connecting to existing systems)
  • Evaluation = The inspection (making sure everything works)
  • Security = The locks and alarms (keeping things safe)
  • MCP = The standard electrical outlets (easy connections)

How These Concepts Differ​

Vector Databases vs. Knowledge Bases​

Vector Databases:

  • Focus: Storage and retrieval of embeddings
  • Purpose: Fast semantic search
  • Scope: Technical infrastructure component
  • Examples: Pinecone, Weaviate, Chroma

Knowledge Bases:

  • Focus: Document processing, chunking, and retrieval logic
  • Purpose: Comprehensive knowledge management
  • Scope: End-to-end RAG solution
  • Examples: Amazon Bedrock Knowledge Bases, Custom solutions

Key Difference: Vector databases are infrastructure; knowledge bases are complete solutions.

Agent Frameworks vs. Tool Systems​

Agent Frameworks:

  • Focus: Overall agent infrastructure and orchestration
  • Purpose: Agent development and management
  • Scope: Complete agent lifecycle
  • Examples: Strands Agents, LangGraph, CrewAI

Tool Systems:

  • Focus: External system integration
  • Purpose: Tool access and execution
  • Scope: Tool-specific functionality
  • Examples: MCP Servers, API integrations, Database connectors

Key Difference: Frameworks are broader; tool systems are more focused.

Memory Systems vs. Knowledge Bases​

Memory Systems:

  • Focus: Conversation context and user preferences
  • Purpose: Personalized AI experiences
  • Scope: User-specific data
  • Examples: AgentCore Memory, Redis, DynamoDB

Knowledge Bases:

  • Focus: Domain knowledge and factual information
  • Purpose: Accurate information retrieval
  • Scope: Organization-wide data
  • Examples: Document stores, Vector databases, RAG systems

Key Difference: Memory is personal; knowledge is organizational.

Model Serving vs. Agent Frameworks​

Model Serving:

  • Focus: Running foundation models efficiently
  • Purpose: Model execution and scaling
  • Scope: Infrastructure for model hosting
  • Examples: Amazon Bedrock, SageMaker, Azure OpenAI

Agent Frameworks:

  • Focus: Agent-specific capabilities and orchestration
  • Purpose: Agent development and management
  • Scope: Complete agent functionality
  • Examples: Strands Agents, LangGraph, CrewAI

Key Difference: Model serving is infrastructure; agent frameworks are application layer.

Example Agentic Applications​

  • Text Generation: Articles, code, documentation
  • Image Generation: Visual content, design assets
  • Audio/Video: Voice synthesis, video creation
  • Code Generation: Programming assistance, automation

Next Steps​

Ready to dive deeper? Check out:

This guide gives you the foundation you need to understand and work with GenAI systems. Each concept builds on the others, so understanding all of them helps you make better decisions about your AI projects.

🤖 AI Metadata
# AI METADATA - DO NOT REMOVE OR MODIFY
# AI_UPDATE_INSTRUCTIONS:
# This document should be updated when new agentic system fundamentals emerge,
# core concepts evolve, or foundational technologies change.
#
# 1. SCAN_SOURCES: Monitor agentic AI research, foundation model updates,
#    core technology developments, and fundamental concept evolution
# 2. EXTRACT_DATA: Extract new fundamental concepts, core technologies,
#    component relationships, and foundational patterns from authoritative sources
# 3. UPDATE_CONTENT: Add new fundamentals, update component descriptions,
#    and ensure all foundational concepts remain current and relevant
# 4. VERIFY_CHANGES: Cross-reference new content with multiple sources and ensure
#    consistency with existing fundamental concepts and component relationships
# 5. MAINTAIN_FORMAT: Preserve the structured format with clear concept descriptions,
#    component relationships, and foundational understanding
#
# CONTENT_PATTERNS:
# - Core Concepts: Direct Prompting, RAG, Memory Management, Vector Stores, Infrastructure as Code
# - Component Relationships: How different agentic system components work together
# - Key Technologies: Language models, vector databases, agent frameworks, deployment patterns
# - Core Questions: Educational questions for understanding agentic system fundamentals
#
# DATA_SOURCES:
# - Agentic AI Research: Foundation models, core concepts, fundamental technologies
# - Component Analysis: Agentic system components and their roles and relationships
# - Technology Landscape: Language models, vector databases, agent frameworks, IaC
# - Additional Resources: Core agentic system concepts, foundational technologies
#
# RESEARCH_STATUS:
# - Fundamentals: Core agentic system concepts and component relationships documented
# - Educational Focus: Content structured for understanding agentic system fundamentals
# - Component Analysis: Comprehensive agentic system components and their roles documented
# - Blog Post Structure: Adheres to /prompts/author/blog-post-structure.md
# - Content Integration: Merged from multiple sources (domain overview, intermediate concepts)
# - Document Evolution: Transformed from documentation to comprehensive blog post
#
# CONTENT_SECTIONS:
# 1. Purpose Section (Purpose of Guide, Core Questions, When to Use, What You'll Learn)
# 2. Core Concepts (Foundation Models, Prompting, RAG, Vector Databases, Memory Systems, etc.)
# 3. Example Applications (Content Creation, Business Applications, Technical Infrastructure)
# 4. Component Relationships (How components work together)
# 5. Practical Considerations (Implementation, development, future, challenges)
# 6. Next Steps (Design decision guidance)
#
# FUNDAMENTAL_CONCEPTS:
# - Direct Prompting: Simple, effective, limited to model knowledge
# - RAG: Retrieval Augmented Generation, combines retrieval with generation
# - Memory Management: Conversation context and user preferences
# - Vector Stores: Semantic search and similarity matching
# - Agent Frameworks: Autonomous AI agent infrastructure
# - Infrastructure as Code: Automated, reproducible agentic system deployments
#
# STRUCTURING_APPROACH:
# - Interactive Learning Diagram: High-level Mermaid flowchart showing concept relationships
# - Clickable Navigation: Direct links from diagram nodes to detailed sections
# - Tabbed Content: Each concept has Definition, Types, Implementations, Differentiating Factors
# - Two-Column Comparisons: Side-by-side layout for "How These Concepts Differ" section
# - Visual Hierarchy: Clear progression from fundamentals to differences to next steps
# - Emoji Integration: Visual cues with italicized angle-bracket formatting (*<<emoji description>>*)
# - Color-Coded Categories: Distinct colors for Core Concepts, Memory, Action, Quality
# - Clean Background: Light gray/white background with colored borders for distinction
#
# ITERATION_HISTORY:
# - Initial: Comprehensive glossary with tabbed sections for each concept
# - Enhanced: Added interactive Mermaid diagram with clickable links
# - Refined: Simplified diagram to focus on core concepts and relationships
# - Improved: Added two-column layout for concept comparisons
# - Polished: Clean color scheme with light background and distinct borders
# - Final: Interactive diagram with direct navigation, no redundant text links
#
# DIAGRAM_STRUCTURE:
# - High-Level Overview: Shows GenAI fundamentals and their relationships
# - Category Grouping: Core Concepts, Memory & Context, Action & Integration, Quality & Safety
# - Relationship Mapping: Solid lines for direct relationships, dotted for cross-connections
# - Interactive Elements: Clickable nodes that link to detailed sections
# - Visual Design: Clean, professional appearance with subtle background colors
# - Navigation Integration: Seamless connection between diagram and content sections
#
# BLOG_POST_EVOLUTION:
# - Original Purpose: Documentation-style glossary of GenAI fundamentals
# - Content Integration: Merged domain overview and intermediate concepts
# - Structure Enhancement: Added tabbed purpose section and example applications
# - Visual Improvements: Interactive Mermaid diagram with clickable navigation
# - Content Consolidation: Removed duplicate sections and integrated advanced patterns
# - Final Form: Comprehensive blog post covering fundamentals through applications
#
# INTEGRATION_HISTORY:
# - Domain Overview: Merged Figma embed and application examples from understanding-genai-domain.md
# - Intermediate Concepts: Integrated advanced RAG patterns and deployment strategies
# - Content Organization: Properly categorized Infrastructure as Code as separate concept
# - Pattern Integration: Moved advanced patterns into appropriate existing sections
# - Structure Optimization: Converted Key Applications to tabbed Example Agentic Applications