Agentic Apps: How Big Companies Are Actually Deploying AI Agents

YouTube · Watch/Read · (gathered 2026-10-14)

• The core argument is to move beyond single-agent solutions and build “Agentic Apps” (Oracle’s definition), which are holistic applications focused on a specific business outcome, integrating multiple AI agents as tools within a larger, deterministic workflow.
• A key concept is that any LLM invocation within an Agentic App must be part of a larger, deterministic workflow. This ensures reliability and control through structured processes including code nodes, conditionals, and context injection, contrasting with less controlled direct LLM interactions.
• Agentic Apps enable enhanced security and reliability through “policies” that define granular, codified agent capabilities. For critical information (e.g., HR policies), knowledge should be codified into tested functions for deterministic, 100% accurate answers, rather than relying on probabilistic RAG.
• A practical takeaway is to always prioritize the business outcome first when integrating agentic AI to ensure value delivery. Additionally, Agentic Apps should leverage visual outputs (e.g., PDFs, diagrams) from agents to improve user comprehension and reduce mental fatigue, moving beyond text-only responses.