What is agentic AI today, and what do we want it to be? (news.mit.edu)

The rapid deployment of AI agents is reshaping business landscapes, with a recent MIT Sloan and Boston Consulting Group report revealing that 35% of surveyed companies have already implemented these systems. Agentic AI distinguishes itself from generative models like ChatGPT by enabling action-oriented functionality, allowing for tasks such as booking flights or interacting with physical devices. This technology is built on foundational generative AI systems but is adapted with specific tools to fulfill particular applications, enhancing user interactions across various platforms. Despite its promising applications, particularly in coding, the development of agentic AI faces challenges, primarily due to limited training data and the complexity of task modeling. As AI agents learn through trial and error, there are risks associated with reliance on these systems, including the potential for errors, the leakage of private data, and the risk of de-skilling users. Looking ahead, the journey of agentic AI raises essential questions about integrating diverse data modalities and architecture advancements to create more powerful and intelligent systems, marking a critical juncture in the evolution of AI capabilities.