Build Systems, Not Code - Angie Jones, Agentic AI Foundation

Building effective agentic systems requires applying traditional software engineering disciplines rather than relying solely on large, monolithic prompts. Systems thinking and workflow design ensure agents function as reliable components with defined boundaries, dependencies, and failure modes. Decomposition and modularity—such as utilizing sub-agents and reusable skills—prevent complexity bloat and improve maintainability. Algorithmic thinking helps distinguish between tasks requiring deterministic code and those needing the model's judgment, while structured output contracts enable safe, automated handoffs between system parts. Designing for idempotency and implementing rigorous threat modeling further stabilize these systems against messy real-world inputs. Ultimately, treating agents as modular, well-documented components allows developers to scale complex workflows while maintaining the same engineering rigor used in pre-generative AI software development, as demonstrated by the design of a house-hunting agent, RelocationScout.