Agentic AI Foundations

Agentic AI Foundations

Agentic AI Foundations is Kavriq’s 12-part core curriculum for learning the building blocks of modern agent systems.

This series starts from first principles and moves toward advanced concepts: agent loops, cognitive architecture, planning, tools, memory, multi-agent systems, guardrails, evaluation, runtime internals, and capstone projects.

If you want the systems-first production perspective after learning the components, continue into Engineering Agent Systems.

This is a living roadmap. Articles are written and published one by one. Items listed without a link are planned and coming. Linked items are published and ready to read.

Prerequisite

• Setting up Environment

Module 1 — Core Concepts

The cognitive and computational foundations of modern agent systems.

• What is an Agent?
• Cognitive Architecture of Agents
• The Inference-Time Compute Revolution
• Modern LLM Primitives

Module 2 — Internal Agent Architecture

The internal components that make an AI system behave like an autonomous agent.

• The Anatomy of an Agent
• The Perception Layer
• Working Memory and the Scratchpad
• The Planner / Reasoner
• The Tool Manager
• The Execution Engine
• The Observation Processor
• Reflection and Termination

Module 3 — Planning Systems

Techniques that allow agents to solve complex tasks through multi-step reasoning.

• Why Planning Matters
• ReAct: Reason + Act
• Chain-of-Thought Planning
• Tree-of-Thought Reasoning
• Execution Graphs
• Building Agents with LangGraph - Python

Module 4 — Tool Use & Protocols

How agents interact with APIs, databases, and external systems.

• Why Tools Make Agents Powerful
• Designing Reliable Tools
• The Model Context Protocol (MCP)

Module 5 — Memory Systems & RAG

How agents store knowledge and retrieve information across interactions.

• The Memory Hierarchy of Agents
• Episodic Memory
• Semantic Memory
• Procedural Memory
• Agentic RAG
• Multi-Hop Retrieval

Module 6 — Multi-Agent Systems

Architectures where multiple agents collaborate to solve problems.

• Why Multi-Agent Systems Exist
• Manager-Worker Coordination
• Handoff Pattern (Swarm)
• Debate Pattern
• Agent-to-Agent Communication (A2A)

Module 7 — Computer Use & Vision

Agents that interact with software interfaces and visual environments.

• Computer Use Agents
• GUI Navigation
• Visual Grounding

Module 8 — Guardrails & Safety

Designing safe and reliable agent systems.

• Prompt Injection Attacks
• Tool Permission Systems
• Human-in-the-Loop
• Sandboxing Agent Execution

Module 9 — Evaluation & Metrics

How to measure agent performance and reliability.

• Why Agent Evaluation Is Hard
• LLM-as-a-Judge
• Trajectory Evaluation
• Building Evaluation Pipelines

Module 10 — High-Performance Engineering

Engineering techniques for scalable, high-performance agent systems.

• The Small Model Strategy
• Observability for Agents

Module 11 — Agent Internals

Understanding how agents actually work by building a minimal runtime from scratch.

• Why Build Your Own Agent Runtime
• Designing a Simple Agent State Machine
• Implementing Tool Calling & MCP Integration
• Adding Time-Travel Debugging
• A Production-Ready 300-Line Agent Runtime

Module 12 — Capstone Projects

Real-world applications built using agentic architectures.

• The Computer-Use Researcher
• The Multi-Agent Coding Pipeline
• The Privacy-First Local Butler

What You Will Learn

By the end of this guide you will understand how to build:

• Autonomous AI agents
• Tool-using reasoning systems
• Multi-agent collaboration architectures
• Production-grade agent infrastructure
• Privacy-first local AI assistants

This series provides a complete technical foundation for modern agentic AI systems.