Skip to main content

Module 3 — Agent Architecture

What this module is about

In Module 2, you built the semantic foundation: an ontology (what exists), a constraint matrix (what rules apply), and semantic contracts (what each component receives, decides, and produces). These artifacts define the WHAT — what the system knows and how it should behave.

This module answers the next question: WHO does the work?

In a traditional software system, you write functions and assign them to services. In an AI system, you design agents — intelligent components that receive context, make decisions, and produce outputs according to their contracts.

But a single agent rarely does everything. Just as a hospital doesn't have one doctor who handles everything from pediatrics to surgery, an AI system has multiple agents, each specialized in a role. The SSA's job is to design this team: who does what, how they coordinate, and what happens when things go wrong.


Learning outcomes

By the end of this module, you will be able to:

  1. Design agent roles with clear boundaries — each agent knows what it's responsible for and what it must never do
  2. Choose orchestration topologies — decide whether agents work under a supervisor, in a pipeline, or as a coordinated swarm, based on the requirements of each system
  3. Define handoff protocols — specify exactly what information passes between agents, what triggers an escalation, and how failures are detected and recovered
  4. Anticipate failure modes — identify what can go wrong in multi-agent systems and design recovery mechanisms before problems occur

Module map

#ContentTypeEstimated time
1Role design — cognitive specializationLesson90 min
2Orchestration topologiesLesson90 min
3Handoff protocols and safetyLesson90 min
AMulti-agent design packageApplication3 hours
BFailure simulation drillsApplication2 hours
AssessmentEvaluation30 min

What you will build

  • An agent matrix defining 6+ agents with roles, scopes, and boundaries
  • A topology decision memo comparing three orchestration approaches and justifying your choice
  • A handoff protocol specification with message schemas, confidence signals, and escalation criteria
  • A failure drill report documenting 5+ failure scenarios with detection and recovery strategies

Prerequisites

  • Module 2 completed (you need an ontology, constraint matrix, and semantic contracts to design agents that use them)
  • Understanding of what semantic contracts specify (input, decision, output)

Estimated time

10-12 hours total (lessons + applications + assessment)