Generative AI is evolving from code generation toward agents that plan, act, evaluate, and improve their work in pursuit of a goal. This shift is beginning to change both software development and the way enterprises organize work.

Drawing on firsthand development experience, this talk traces three stages of AI-driven development, highlighting productivity gains alongside the challenges of maintaining architectural consistency and code quality. It then explains how goal-driven agentic loops work and explores implementation approaches that enable multiple agents to share responsibilities.

An execution environment combining local models and external AI services serves as one implementation example. Through this example, the talk examines how model choice, data access controls, human oversight, and enterprise governance can be incorporated into agent systems. Building on lessons from software development, the discussion looks ahead to cross-functional coordination and compliance activities that have been difficult to automate, presenting a vision of enterprise operations in which people and AI agents work together.