I see RAG and Agentic AI as complementary: RAG helps AI find the right information, while Agentic AI helps AI decide what to do next. I do not think Agentic AI will replace RAG; the strongest approach is to combine them.
My approach is: Start with RAG; add agents when a task genuinely requires multiple decisions or actions.
For an AI support agent, I would retrieve company policies, check order details, and follow a workflow, with human approval for critical actions. This illustrates how retrieval and agent capabilities can work together while keeping important actions under human oversight.
I prioritize production reliability over autonomy alone. The post’s goal is not to build increasingly complex AI, but to build useful and reliable AI.
My goal is building more useful and reliable AI.