AI Architecture & Agentic Systems Training

AI Architecture & Agentic Systems

• From AI Tools to AI Actors
  • Limitations of chat-based AI systems
  • Why autonomy changes system risk
  • The shift from “response generation” to “action execution”
  • Real-world examples of agentic behavior
• What Makes a System Agentic
  • Defining autonomy, goals, and decision loops
  • Planning vs execution
  • Internal vs external control
  • Degrees of autonomy and supervision
• Core Components of Agentic Architectures
  • The agent core (reasoning and planning)
  • Tools, actions, and external systems
  • Memory and state management
  • Orchestration and coordination
• Authority, Boundaries, and Scope
  • Defining what an agent is allowed to do
  • Preventing overreach and unintended actions
  • Scoping authority to roles and tasks
  • Designing safe defaults
• Memory, Context, and State
  • Short-term vs long-term memory
  • State persistence and recovery
  • Context drift and contamination
  • Strategies for memory isolation
• Control Loops and Supervision
  • Planning, acting, observing, and adjusting
  • Human-in-the-loop vs human-on-the-loop
  • Monitoring agent behavior
  • Interrupting and stopping agents safely
• Common Failure Modes in Agentic Systems
  • Runaway behavior and infinite loops
  • Hallucinated plans and actions
  • Tool misuse and incorrect assumptions
  • Compounding errors across steps
• Designing for Safety and Reliability
  • Containment and blast-radius reduction
  • Idempotency and repeatability
  • Auditability and traceability
  • Aligning agent behavior with governance
• Multi-Agent and Orchestrated Systems
  • When multiple agents are appropriate
  • Coordination and communication patterns
  • Shared vs isolated state
  • Failure propagation across agents
• Evaluating Agentic System Designs
  • Architectural review criteria
  • Risk assessment for autonomy
  • Readiness for implementation
  • Red flags and anti-patterns
• Preparing for Structured Implementation
  • Why ad-hoc agents fail in production
  • The role of protocols and schemas
  • Transitioning to MCP-based systems
  • Preparing for hands-on agent implementation
• Summary and Next Steps
  • Key architectural principles
  • Applying agentic thinking responsibly
  • Aligning architecture with organizational maturity
  • Moving into MCP-based system design and development

Before Taking this Class

Completion of AI Foundations & Risk for IT Professionals and Designing Reliable AI Workflows & Interactions, or equivalent experience designing AI-enabled workflows with an understanding of risk and governance.

Software/Setup For this Class

A laptop or desktop computer with a modern web browser and reliable internet access is recommended for accessing course materials and participating in discussions. No programming tools or AI software access is required.