Building AI agents is 5% AI and 100% software engineering

The article emphasizes that the success of production-grade AI agents hinges primarily on robust data infrastructure, controls, and observability rather than the choice of underlying models. It details a "doc-to-chat" pipeline that ingests, standardizes, and governs enterprise documents, leveraging embedding indexes and relational features to enable secure, auditable retrieval and generation through APIs with human-in-the-loop checkpoints, forming the backbone for agentic Q&A, copilots, and workflow automation. Furthermore, the architecture integrates seamlessly with existing enterprise stacks by employing standard service protocols like REST/JSON and gRPC, utilizing trusted storage solutions such