Agentic AI for Software Engineers — Use Your Engineering Skills to Build Better AI

Software engineering skills transfer directly to agentic AI development: system design maps to context engine architecture, API design maps to MCP tool schema design, testing discipline maps to LLM component testing with mocked responses, observability engineering maps to Glass-Box logging implementation, and production operations maps to AI system deployment and monitoring. The discipline is the same; the domain is new. This workshop shows the mapping at each point.

Software engineers find these agentic AI concepts immediately intuitive: MCP typed interfaces (analogous to API contracts), the Glass-Box observability layer (analogous to distributed tracing), context window budget management (analogous to memory management), semantic blueprint versioning (analogous to schema versioning), and the hub-and-spoke multi-agent pattern (analogous to microservices with an API gateway). The workshop leverages these analogies to accelerate understanding.

The genuinely new concepts for software engineers are: semantic blueprint design (how to structure instructions that reliably guide LLM behavior), RAG pipeline engineering (how embedding-based retrieval works and how to build production retrieval systems), hallucination prevention (how to ground LLM outputs in verified sources), context rot (how LLM performance degrades as context accumulates), and prompt injection (how adversarial text can override agent instructions). The workshop introduces each of these clearly before implementing them.

Software engineering testing discipline applies to agentic AI by treating LLM responses as external dependencies that should be mocked in unit tests, similar to database or API calls. Unit tests verify that the non-LLM logic (context routing, citation parsing, schema validation) works correctly with controlled mock responses. Integration tests verify component interactions with a test LLM that produces predictable outputs. This approach makes agentic AI systems testable and regression-proof in the same way as other complex software systems.

The direct pattern analogies are: the Glass-Box Context Engine is a pipeline architecture (familiar from data engineering), MCP servers are microservices with typed RPC interfaces (familiar from service-oriented architecture), semantic blueprints are structured configuration artifacts (familiar from configuration-driven systems), RAG is a hybrid search system (familiar from search engineering), and episodic memory is an event store (familiar from event-driven architecture). Denis Rothman explicitly draws these analogies throughout the workshop.

Software engineers are typically productive with agentic AI immediately after this workshop because they already have the foundational skills: Python, system design thinking, API design, and testing discipline. The workshop fills the agentic AI-specific knowledge gaps: context engineering patterns, MCP implementation, RAG pipeline design, and production LLM system operations. Most software engineers who attend report being able to meaningfully contribute to agentic AI projects the week after the workshop.