Agentic Apps: How Big Companies Are Actually Deploying AI Agents

This post introduces the concept of 'agentic apps' as a deployment strategy for AI agents in large companies, moving beyond single-agent interactions to multi-agent systems orchestrated within workflows. It emphasizes using AI judiciously for judgment, converting rules into tested code, and incorporating human approval for critical actions. A case study of Oracle AI for Fusion Applications is presented, showcasing agents that prioritize actions and execute LLM calls within workflows, incorporating security policies and visual outputs. The talk outlines building these apps using tools like Oracle AI Agent Studio, where workflows manage LLM calls, and policies are implemented as nodes within these workflows, distinct from RAG. The discussion covers workflow permissions, triggers, and the importance of visual outputs for user comprehension.