This course contains the use of artificial intelligence.

Build Reliable Agentic AI Systems

Agentic AI is more than an LLM responding to a prompt. A reliable agent operates through a controlled loop: it interprets a goal, selects an action, uses a tool, observes the result, evaluates progress, and continues until it reaches a verified outcome.

This hands-on course teaches the foundations of Loop Engineering for Agentic AI. You will learn how to design, build, control, debug, and evaluate agent loops that perform meaningful work without becoming unpredictable, repetitive, or unsafe.

What You Will Build

You will build one evolving Python project throughout the course. Starting with a minimal tool-using agent, you will progressively add:

• Tool calling and validated action schemas

• State, memory, checkpoints, and recovery

• Context-window management and compaction

• Termination conditions and resource limits

• Guardrails and permission boundaries

• Tracing, verification, and debugging

• Multi-agent orchestration and handoffs

• Human approval checkpoints

The final capstone is a reliable issue-resolution agent that can inspect a repository, use development tools, preserve progress, detect non-progress, delegate verification, request approval, and produce an auditable execution report.

What You Will Learn

• Explain how an agentic loop differs from a single LLM call

• Design the goal–act–observe–evaluate cycle

• Build a working tool-calling agent loop in Python

• Create clear tool contracts and validate agent actions

• Handle tool errors, retries, timeouts, and invalid requests

• Manage state and memory across agent iterations

• Checkpoint, resume, and recover interrupted agent runs

• Control context growth and prevent context drift

• Define reliable success, failure, blocked, and escalation outcomes

• Detect repetition, oscillation, and non-progress

• Apply permissions, guardrails, and risk-based approvals

• Verify outcomes using tests, validators, and reviewer agents

• Trace, replay, diagnose, and repair failed runs

• Implement sub-agent, orchestrator-worker, and handoff patterns

• Apply Loop Engineering concepts with Claude Code

• Prepare agentic systems for safe production use

Claude Code and Multi-Agent Workflows

A dedicated section demonstrates how Claude Code can support Loop Engineering through project instructions, skills, plugins, MCP integrations, hooks, automations, specialized sub-agents, permissions, and Git worktrees.

You will compare single-agent and multi-agent designs, implement an orchestrator-worker workflow, define reliable handoff contracts, and prevent delegation loops or conflicting work.

Hands-On Course Format

• Concise, focused theory

• Progressive guided labs

• Python coding exercises

• Decision-based role-play activities

• Section quizzes

• Two full-length practice tests

• Reusable templates and checklists

• One integrated capstone project

A mock LLM adapter supports no-cost practice. An optional live-model adapter is included for learners who want to experiment with a real model provider.

Who This Course Is For

• AI engineers building agentic applications

• Software developers moving beyond basic prompting

• Solution architects designing reliable AI systems

• Technical leads evaluating agent architectures

• Automation engineers creating tool-driven workflows

• Learners interested in Claude Code and multi-agent development

Basic Python knowledge is helpful, but prior experience building AI agents is not required.

Course Outcome

By the end of the course, you will understand not only how to make an agent act, but also how to make it stop correctly, recover safely, verify completion, escalate intelligently, and remain under meaningful human control.