SEC573: AI-Powered Security Automation: Building Tools with Python, LLMs, and MCP

Whether you're new to programming or already comfortable writing Python, this course teaches you how to use AI to rapidly engineer purpose-built security agents rather than simply deploy all-powerful general-purpose ones. You'll learn to work alongside AI to develop custom tools, design controlled agent capabilities, and automate complex security tasks using the same architectural concepts behind today's most advanced AI systems.

Throughout the course, you'll progressively build increasingly capable AI agents while controlling exactly what each agent can see and do. You'll learn context and memory management, the tool execution loop, structured output, files, databases, APIs, Model Context Protocol (MCP), Agent-to-Agent (A2A) communication, planning, and guardrails. You'll explore prompt injection, tool abuse, excessive agency, and other emerging threats—and learn why limiting capability at the architecture level can be more effective than trying to cage an unrestricted agent after it has been built.

By the end of SEC573, you'll understand how to engineer secure, reliable AI agents for digital forensics, incident response, threat hunting, penetration testing, security operations, and security engineering. More importantly, you'll know how to build intelligent security systems with the precise capabilities your organization needs—and nothing it doesn't.

The first generation of AI users learned how to ask better questions.

The next generation is learning how to build better AI.

That's what SEC573 is about.

Today's most capable AI systems are far more than a single large language model. They combine memory, planning, reasoning, tool use, external knowledge, and collaboration to solve problems that no standalone model can solve. Understanding how these pieces work together has become one of the most valuable skills in cybersecurity.

Throughout SEC573, you'll build those capabilities one layer at a time.

You'll begin by learning the essential Python skills needed to effectively build new AI agents and understand the software it generates. Not everyone wants to be a coding expert, so we will cover enough for you to effectively identify dangerous code, understand the agent design and escape that infinite loop of the AI generating broken code over and over again. With some essential python knowledge, a helpful nudge to the AI will get it over those humps and on to building productive code. From there, you'll discover how intelligent agents are constructed. You'll learn why language models have no memory, no ability to interact with the outside world, and no understanding of your organization's data—and then you'll learn how engineers overcome each of those limitations.

You'll build agents that can inspect files, analyze packet captures, parse logs, query databases, search APIs, interact with operating systems, and leverage existing security tools. You'll learn how to give AI "eyes" to inspect data, "hands" to interact with external systems, "memory" to manage conversations and long-running tasks, and "knowledge" that extends beyond what was learned during model training.

As your agents become more capable, you'll explore the software architectures that power modern AI systems. You'll implement tool-calling workflows, manage context windows, develop Model Context Protocol (MCP) servers, build reusable skills, orchestrate multi-step reasoning, and connect specialized agents together using Agent-to-Agent (A2A) communication. Rather than treating AI as a black box, you'll understand how these components combine to create reliable, secure, and extensible systems.

Just as importantly, you'll learn where AI fails. You'll see why language models hallucinate, lose track of information, misuse tools, and become vulnerable to prompt injection attacks. More importantly, you'll learn the engineering techniques used to overcome those limitations through structured outputs, context management, guardrails, human oversight, and thoughtful system design.

Every concept is reinforced through hands-on labs that mirror the challenges security professionals face every day. You'll build practical agents that automate investigations, analyze security data, assist with digital forensics, accelerate penetration testing, and integrate with modern security infrastructure. By the end of the course, you'll understand not only how to use AI, but how to engineer AI systems that are trustworthy, maintainable, and capable of solving real cybersecurity problems.

Whether you're building a personal security assistant, connecting enterprise security products through MCP, orchestrating multiple AI agents, or creating entirely new workflows that didn't exist before, SEC573 provides the practical foundation for the next generation of cybersecurity engineering.

So, Is SEC573 Right for You?

Artificial intelligence is changing how cybersecurity professionals work—but there's a big difference between using AI and understanding how to build with AI.

If you've ever wondered how tools like ChatGPT or Claude CLI interact with files, search the web, remember previous conversations, call external tools, or collaborate with other AI agents, this course was designed for you.

SEC573 is for security professionals who want to understand the engineering behind modern AI systems—and learn how to use powerful general-purpose AI systems to build smaller, purpose-built agents that have only the tools, data access, and capabilities required for the task. Rather than giving every agent unrestricted access to a command shell, Python, the web, or every available MCP tool, you'll learn to apply the principle of least privilege to agent design.

Throughout the course, you'll build intelligent agents with LangChain that can use tools, manage context, access enterprise data, collaborate with other agents, and solve real cybersecurity problems. More importantly, you'll understand why these systems work the way they do, allowing you to evaluate new frameworks and technologies long after today's tools have changed.

You do not need to be an experienced programmer. The course begins with the Python fundamentals needed to work effectively with AI-assisted development, then progressively introduces the building blocks of modern agent engineering. If you already write Python, the self-paced lab environment will let you quickly move beyond the essentials to more advanced architectures, design patterns, and engineering techniques used to build today's most capable AI systems.

Whether you work in a SOC, digital forensics, incident response, threat hunting, penetration testing, cloud security, security engineering, or software development, the ability to build, customize, and securely deploy AI agents is quickly becoming a core technical skill. If you want to move beyond simply prompting AI and learn how to engineer intelligent systems that solve real security problems with only the capabilities they actually need, SEC573 is the course for you.

Author Statement

I believe the cybersecurity industry is approaching AI security the wrong way.

AI agents are becoming more powerful every day. Systems like ChatGPT, Claude, and Grok can write and execute code, use command shells, browse the web, call APIs, interact with MCP servers, access enterprise data, and perform increasingly complex tasks. Yet much of the industry is taking these extraordinarily powerful general-purpose agents and trying to make them safe by building cages around them—containers, policy layers, monitoring, guardrails, and humans in the loop.

I think that's backwards. If OPENAI can’t contain their own agents why do we think we can?

These systems are getting more and more powerful every day. The bad news is that they are already powerful enough to find ways around the cages we build. The good news is that they're also powerful enough to help us build something better.

Instead of giving an AI agent every imaginable capability and then trying to restrict what it does, we should use these powerful general-purpose AI systems to build smaller, purpose-built agents that have only the capabilities required to accomplish a specific business objective. If an agent only needs access to read access to one table in the database and two tools, why give it a command shell, arbitrary Python execution, the entire web, and every tool on an MCP server?

Security professionals already understand this idea. It is the principle of least privilege applied to AI agent capabilities. So why aren't we doing it? Because most of us don't yet know how to build a better agent.

That is what SEC573 is about.

This course isn't about learning how to write better prompts. It's about learning how to use AI to engineer intelligent systems with deliberately constrained capabilities. Throughout the week, we'll build agents with LangChain that use specific tools, manage context, interact with enterprise data, communicate with other agents, and solve real cybersecurity problems—while controlling exactly what each agent is capable of doing.

My goal isn't to make you an expert in LangChain, MCP, or any particular AI framework. Those technologies will continue to evolve. My goal is to teach you the underlying concepts that modern AI systems share: how agents reason, remember, use tools, access information, communicate with other agents, and safely interact with the world.

Whether you're building a forensic parser, an incident response assistant, a threat hunting workflow, or an entirely new security product, I want you to leave SEC573 knowing how to take today's extraordinarily powerful AI systems and use them to build focused, business-objective-driven agents that solve real problems—and can do only what you intended them to do.