The deployment of automated software systems known as agentic AI has surged recently. According to a November 2025 report from MIT Sloan School of Management and Boston Consulting Group, 35% of surveyed businesses have already deployed AI agents, while another 44% plan to do so soon.
What is Agentic AI? Agentic AI refers to AI that takes actions in the world, either physically (like robotic manipulation) or digitally (like booking a flight). In contrast, generative AI focuses on creating text, art, and stories rather than performing tasks. The term 'agent' usually denotes AI that assists users in interacting with applications, websites, or the physical world. Most agents encountered today are digital, such as customer service bots handling product complaints.
Building Agentic AI Most companies use similar AI models as a foundation, enabling them to take actions and remember past events. An agent typically starts with a generative AI system at its core, with additional tools tailored for specific applications. For instance, an agent might have access to a calculator for solving math problems or a more complex system for recalling financial data and past negotiations.
Challenges in Training Agentic AI The biggest challenge in developing agentic AI is the lack of training data. For example, creating a system that can book flights online seems simple, but there isn't enough data detailing the exact steps. One training method involves having the AI agent visit airline websites, experiment, and learn what works.
Promising Applications of Agentic AI The coding domain has seen significant success with agentic AI, evolving from generative AI. Language models trained on code can predict how humans would solve coding problems. Agents can learn through feedback loops, trying various solutions and verifying correctness. However, balancing automation with human assistance remains crucial, especially in high-stakes situations.
Risks of Using AI Agents A major risk is that the ease of using AI agents may lead users to neglect verification of the output. This can introduce bugs and lead to data leaks. Additionally, reliance on agents may result in de-skilling, where individuals lose the ability to perform tasks independently.
Future of Agentic AI Currently, agentic AI mainly involves large language models interacting with tools. Future advancements may require models capable of handling video, physical forces, time series, and other complex data. The question remains whether future AI will evolve from existing models or be built from entirely new architectures that can better understand and interact with the world.
Blogger's Review: The rise of agentic AI in automation invokes profound considerations. While its potential is vast, it’s crucial to recognize the importance of security and ethical responsibility. Only through a comprehensive understanding and mastery of these tools can we achieve sustainable intelligent development.