The use of automated software systems, known as AI agents, has surged recently. A November 2025 report from the MIT Sloan School of Management and Boston Consulting Group revealed that 35 percent of companies surveyed had already deployed AI agents, with another 44 percent planning to implement agent AI soon.
Understanding Agentic AI
What is Agentic AI?
Agentic AI refers to artificial intelligence that takes action in the world, whether through physical actions like robotic manipulation or digital actions such as booking a flight. Unlike generative AI models like ChatGPT and Claude, which create stories, poems, art, and images, agentic AI is designed to perform tasks.
The term “agent” is essentially a brand name. Typically, this involves AI that assists users in interacting with an application, a website, or the physical world. Most of the agents we encounter today are digital, such as virtual customer service representatives who handle product complaints. These agents are often driven by similar underlying AI models, allowing them to execute actions and retain memory of past interactions.
Challenges in Developing Agent AI
A significant challenge in developing agent AI is the scarcity of training data. For instance, creating a system capable of booking a flight online involves numerous steps, like moving the mouse, clicking buttons, and handling errors, for which comprehensive data is often lacking. One approach to training such systems is through trial and error, where the AI agent interacts with airline websites to discover effective strategies.
Applications and Risks of Agentic AI
Promising Applications
Agentic AI has found success in coding applications. By training language models on code, AI can predict human solutions to coding problems. These agents engage in feedback loops, trying different solutions until they verify the correct answer. However, the balance between automating decision-making and supporting human decision-makers is crucial, especially in high-stakes areas like medicine, security, and business policy.
Risks Involved
One major risk is over-reliance on agents for tasks like coding, which could lead to insufficient oversight and error-checking, resulting in bugs and data leaks. Additionally, improper use or vague instructions from humans can cause the AI to make mistakes. There’s also the concern of deskilling, as people may lose the ability to perform tasks independently if they rely too heavily on AI agents.
The Future of Agentic AI
What’s Next?
Currently, “agentic AI” involves large language models that use tools to interact with digital and physical systems, but these models are primarily based on text data. Future development may require models that handle various data types, including video, physical forces, and more complex modalities. There’s an ongoing debate within the AI community about whether future AI will be an extension of existing models like Claude with added sensors and actuators, or entirely new systems built from the ground up.
For more detailed insights, refer to the original source Here.
“`
