AGENTS & AUTONOMY
Agentic AI
Understand how AI systems move beyond one-shot responses into planning, tool use, memory, feedback and autonomous execution.
Agentic AI Learning Path
Follow the modules in order to build a solid mental model of LLM agents, the agent loop, tools, context, autonomy, safety and evaluation in real systems.
What Is Agentic AI?
The core mental model behind AI agents — why an LLM that can only answer questions eventually stops being enough, and what changes when a system starts deciding and acting on its own.
Why Did AI Agents Become Necessary?
Tracing the real engineering evolution from rule-based software to autonomous agents — the specific limitation that forced each step, and the specific new problem each step created in exchange.
Anatomy of an AI Agent
Every real architectural component of an AI agent, named precisely — goal, model, instructions, tools, environment, state, memory, reasoning, planning, observation, feedback, action, and termination — and how they fit together.
The Agent Loop
A full, working run of the agent loop from goal to resolution — plus what happens when a tool fails, a diagnosis is wrong, information is missing, or permission is denied.
Tools and Tool Calling
How a tool schema works, how the model's decision becomes a real action, why argument validation matters, and what separates a well-designed tool from a poorly-designed one.
Planning, Reasoning and Decision Making
What reasoning and planning are inside an agent, the real difference between a model's own reasoning capability and the orchestration logic built around it, and when planning earns its keep versus when it's unnecessary overhead.
Context, State and Memory
The three-way distinction that trips up more people building agents than almost anything else — what context, state, and memory each are, why an agent needs all three, and how they're different.
Agentic Workflows
The full spectrum from a rigid, fixed workflow to a fully autonomous agent — what each point looks like, and the practical question that matters more than any of them: when should you use a workflow instead of an agent?
Agent Autonomy and Human-in-the-Loop
Autonomy as a spectrum, not a switch — human-in-the-loop, on-the-loop, and out-of-the-loop, why more autonomy means more potential value and more potential risk, and when approval should be mandatory.
Agent Failure and Reliability
How agents fail — hallucinated arguments, wrong tool selection, infinite loops, cascading failures, and more — with what happened, why, how to detect it, how to prevent it, and how to recover, for each one.
Agent Security and Guardrails
What happens when an agent isn't merely mistaken but deliberately manipulated — prompt injection, malicious content, excessive permissions — and the guardrails, enforced in code rather than prompts, that stop it.
Evaluating, Observing and Controlling Agents
Why 'the agent worked once' is not a production metric — what needs to be logged and traced, the metrics that matter, and why testing an agent requires evaluation, not assertions.
Real-World Agentic AI
How companies across law, healthcare, cybersecurity, retail, and customer support are deploying agentic AI in production — what worked, what didn't, and the engineering lesson each real example teaches.
Single-Agent vs. Multi-Agent
When a single, well-designed agent stops being enough, what changes when several agents work together, and the real cost of that specialization — a conceptual bridge, not a deep dive into coordination itself.
When NOT to Use Agentic AI
The final module: a practical decision framework for when traditional software beats an agent, and this course's closing principle — use the least autonomous architecture that reliably solves the problem.
WHY AGENTIC AI?
Move from answering to deciding and acting
Agentic AI adds the missing ingredients for real task completion: the ability to plan, call tools, maintain context, recover from failure and operate with guardrails. The result is a system that can handle multi-step work more like an autonomous teammate than a simple chatbot.