Agentic AI: A New Era of Autonomous Artificial Intelligence Based on OpenAI & DeepMind Research

SAN FRANCISCO, KOMPAS.com — The artificial intelligence industry is undergoing its biggest paradigm shift since the launch of large language models (LLMs). OpenAI and Google DeepMind sequentially published their latest research on Agentic AI—an intelligence system that no longer merely answers questions passively like regular chatbots, but is capable of planning, dividing tasks, and executing complex instructions autonomously without human intervention.

Based on reports from international scientific journals, the fundamental difference of Agentic AI lies in the integration of internal feedback loops and the ability to use external tools such as APIs, web browsers, and shell terminals to manipulate data.

According to Dr. Helena Vance, a senior researcher at the AI Research Institute San Francisco, these autonomous agents are designed to work like virtual employees who are given end-goal instructions, and then figure out on their own how to achieve them efficiently.

How Does Agentic AI Work?

Collaborative research from Stanford University shows that modern AI agents use a highly complex three-layer architecture to process decisions:

1. Planning: AI systematically breaks down large goals into smaller steps through Chain of Thought (CoT) techniques.
2. Cognitive Memory: Uses short-term memory to track current task status and Vector Database-based long-term memory to retrieve past knowledge.
3. Tool Use: The ability to write programming code autonomously, call webhooks, or access web browsers to solve real-world problems.