Secure Agentic AI Engineering: Build Production AI Agents

About this course

This course contains the use of artificial intelligenceAI agents are changing how software is built. An agent does not only talk: it checks systems, calls tools and finishes real tasks. That is also what makes it risky. An agent that can act can delete the wrong data, leak a secret, or be tricked by a single hidden sentence in a document. Companies need engineers who can build agents that are useful AND safe. This course teaches you to be that engineer.

You start by building real AI agents in Python, with the real OpenAI API:Your first agent: a service health agent with one tool, a bounded loop and a live log you read line by line.

A multi-agent team: an investigator agent with tools hands its facts to a writer agent.

An agentic agent: it writes its own plan, does each step with tools, checks its work, then reports.

Then you learn how every part really works, and how to make it safe:

• LLM foundations for engineers: tokens, context windows, temperature, prompts, structured outputs, tool calling and provider adapters.
• The agent loop: state, events, task specs, a closed tool registry, step limits, timeouts and loop prevention.
• Safe tools: argument validation, retries with backoff, idempotency keys, safe HTTP, database and file tools, and human approval gates.
• Agentic reasoning: planning, ReAct, plan-and-execute, reflection, routing, state machines, and when NOT to use an agent.
• Memory and RAG: SQLite memory, embeddings, retrieval, and trust boundaries for retrieved documents.

Multi-agent systems: supervisor and worker agents, handoffs, message contracts, durable workflows and MCP.AI agent security: threat modeling, direct and indirect prompt injection, tool abuse, secret handling, policy enforcement, sandboxing, identity and supply-chain trust.

Testing and reliability:

Prices, discounts and availability are set by Udemy. We may earn a commission when you purchase through links on this site.