Engineering Reliable Agentic AI Systems
A practical, hands-on workshop on System Architecture, Harness Engineering and Loop Engineering
Building an AI agent is easy. Building one that works reliably in production is not.
Many AI systems can complete simple tasks, but they often struggle with longer workflows. They lose context, make unnecessary model calls, fail to verify their output, retry indefinitely, and become increasingly expensive to operate. As AI agents take on more responsibility, prompt engineering alone is no longer enough.
Reliable agentic systems require thoughtful architecture, structured execution, robust evaluation, and observability. They need clear goals, memory, verification, cost-aware decision making, and safe stopping conditions to perform consistently in real-world environments.
This workshop teaches the engineering patterns behind production-ready agentic systems.
Over four hands-on hours, you'll learn how System Architecture provides the foundation, Harness Engineering ensures quality and reliability, and Loop Engineering orchestrates intelligent execution. Through practical examples and guided exercises, you'll explore how these disciplines work together to build AI systems that are scalable, observable, cost-efficient, and trustworthy.
By the End of This Workshop, You'll Be Able To
- Design the architecture for reliable agentic AI systems.
- Structure autonomous agent workflows using modern Loop Engineering principles.
- Write machine-readable specifications using Spec-Driven Development (SDD).
- Build evaluation, testing, and execution harnesses to validate AI agent behaviour.
- Implement deterministic verification, LLM-as-Judge, and Agent-as-Judge evaluation strategies.
- Design agents that make efficient use of context windows and minimize unnecessary token usage.
- Integrate external tools safely using the Model Context Protocol (MCP).
- Build autonomous research workflows that validate sources and synthesize reliable results.
- Design memory, observability, retry strategies, and human-in-the-loop workflows for production systems.
- Apply reusable architectural patterns to real engineering tasks such as PR reviews, CI maintenance, dependency updates, issue routing, and changelog generation.
Who Is This Workshop For?
This workshop is ideal for:
- Software Engineers building AI-powered development workflows
- AI and LLM Engineers designing autonomous agent systems
- Platform and DevOps Engineers automating engineering operations
- Technical Architects building production AI platforms
- Engineering Managers and Tech Leads evaluating agentic AI adoption
- Developers using Claude Code, Codex, Cursor, Gemini, or similar AI coding assistants who want to build reliable, production-ready AI systems
Prerequisites
- Basic familiarity with Claude Code or another AI coding tool
- GitHub account with Actions enabled
- Python 3.11+ for labs
- Claude API key (free tier works for labs 1-3)
What You'll Receive
- 4 hours of live, hands-on instruction with Rick Hightower
- Guided labs and practical exercises you can follow along with
- Workshop recording to revisit the material after the session (if included)
- Downloadable workshop resources (slides, code, lab materials, if included)
- Live Q&A with Rick throughout the workshop
- Certificate of Completion from Packt
Lineup
Rick Hightower
Good to know
Highlights
- 5 hours
- Online
Refund Policy
Location
Online event
Agenda
Module 1: Architecting Reliable Agentic AI Systems
The workshop begins by exploring why prompt engineering alone doesn't scale for production AI systems. You'll learn the architectural foundations of reliable agentic applications, including models, orchestration, tools, memory, verification, and observability. The session introduces the Discover → Act → Verify execution model and the core building blocks of modern agent systems, including goals, scheduling, skills, worktrees, state management, and stopping conditions.
Module 2: Spec-Driven Development & Harness Engineering
Reliable AI agents need more than instructions. They need clear specifications and robust validation. In this module, you'll learn how to write machine-readable specifications using Spec-Driven Development (SDD) and build evaluation, testing, and execution harnesses that ensure agents behave as expected. You'll also explore Maker/Checker architectures, deterministic verification, LLM-as-Judge and Agent-as-Judge evaluation strategies, context management techniques, token optimization, reducing unnecessary retries, and designing cost-efficient evaluation pipelines.
Module 3: Research Loops & MCP in Practice
This module focuses on building AI agents that can safely interact with external tools and gather reliable information. You'll explore autonomous research loops, learn the fundamentals of the Model Context Protocol (MCP), and understand how MCP tool contracts, safe retries, and idempotent tools contribute to reliable execution. The session also covers search, verification, synthesis, source validation, token-efficient research workflows, and cost-aware search strategies for production systems.