As AIOps and collaboration tools pivot toward “agentic” functionality, many teams struggle to move beyond unreliable chat prompts. The challenge isn’t just “calling an LLM”—it’s architecting a system where agents reliably interact with complex domains, respect business logic, and scale alongside existing microservices.

In this session, we explore Agentic Patterns for Domain-Driven Design, drawing on successful implementations from best-of-breed SaaS platforms. We will dissect the three critical layers of a modern agentic stack—Interaction, Orchestration, and Domain—while focusing on the rigorous engineering required to take a project from concept to code.

Key topics include:

The PRD as Prompt Engineering: How to construct a “spec” that drives planning in IDE assistants. We’ll discuss writing requirements that are machine-readable and context-aware to ensure AI-generated code aligns with your architectural standards.

Testing for Quality Attributes: How to move beyond “vibes-based” testing. We will cover technical validation for scalability, identifying tech debt early in AI-generated modules, and the “Token Tax”—how to model cost (COGS) when tokens drive your unit economics.

The Orchestration Plane: A comparison of the Manager-Delegate pattern versus Serial Agent Tasks, and how to manage execution state at scale.

The Domain & Information Layer: Integrating backend microservices via the Model Context Protocol (MCP) and custom RAG to maintain a “clean” architecture.

From MVP to Mainline: Strategies for transferring your agentic project from a rapid-prototype MVP into a normal development cycle, ensuring it inherits the same CI/CD, observability, and security posture as your core platform.

Audience - software designers adding AI to existing products
Required background - basic familiarity with MCP, RAG, agent orchestration, microservices architecture
Session duration about 1 hour with Q&A