Agentic AI, a form of artificial intelligence capable of taking digital or physical actions on behalf of users, has recently seen widespread adoption across industries. Distinguished from generative AI models like ChatGPT and Claude, which primarily produce content such as text or images, agentic AI integrates these generative abilities with operational tools that allow it to act autonomously. This shift is transforming how companies apply AI in real-world scenarios, although challenges and risks remain, according to expert analysis from MIT researchers.
What Happened
A November 2025 report by the MIT Sloan School of Management and Boston Consulting Group revealed that 35 percent of surveyed businesses had already deployed AI agents, with an additional 44 percent planning to implement agentic AI systems soon. These AI agents build upon foundational generative models like Anthropic’s Claude, employing additional “wrappers” or tools tailored to specific tasks such as financial data management, coding assistance, or customer service.
Phillip Isola, associate professor at MIT’s Electrical Engineering and Computer Science Department and a member of CSAIL, explains that developing effective agentic AI remains challenging, largely due to a scarcity of high-quality training data for action-based tasks. For example, teaching an AI agent to autonomously book airline tickets requires extensive trial-and-error learning given the complexity and variability of web interfaces involved.
Key Facts
The key findings and details include:
- The corporate adoption of AI agents has grown rapidly, with MIT Sloan and BCG’s report underscoring a combined 79 percent of companies either using or planning to use these systems.
- Agentic AI systems are typically based on large language models (LLMs) such as Claude, enhanced with specific tools and memory capabilities tailored to their application.
- Coding assistance represents one of the most successful applications, as the AI can iteratively test and improve code by assessing outputs for correctness.
- Risks identified include potential software bugs, data leaks, and human over-reliance that could lead to errors if AI outputs are not thoroughly vetted.
- The field anticipates architectural innovations that may integrate modalities beyond text, such as video, radar, and physical sensor data, to build more potent AI agents in the future.
What This Means
For businesses and users, the rise of agentic AI signals a fundamental shift from passive content generation to active problem-solving and task execution. This development promises enhanced automation of routine or complex workflows, such as coding, customer interaction, or data management, potentially increasing operational efficiency. However, it also introduces new risks. For instance, coding agents can accelerate development but may propagate unnoticed errors if users do not rigorously validate AI-generated results.
Moreover, the reliance on these systems could influence workforce skills, leading to a de-skilling effect where users become less adept in tasks formerly performed manually. This raises important considerations for workforce training and oversight. Lastly, since current agentic AI models largely depend on textual training data, extending to real-world physical interaction remains nascent and will require significant research advancements.
Background
Agentic AI evolved from generative AI models that create language, code, or images but added capabilities to interact with external tools and environments. Companies have adapted foundational LLMs by equipping them with specific utilities like calculators, file systems, or APIs to execute commands. This trend aligns with broader enterprise efforts to automate processes and streamline decision-making.
What Remains Unclear
The architecture of future agentic AI systems remains an open question within the AI community. Experts debate whether the next generation will be incremental improvements on current LLM-based agents enhanced with multimodal sensors or entirely new models designed from the ground up to process complex, continuous data. Additionally, the timeline and scale for agentic AI to reach widespread physical-world application are still uncertain.
What Comes Next
Research continues at institutions like MIT CSAIL to improve AI agent capabilities and reliability. Meanwhile, businesses are expected to accelerate deployments especially in digital domains such as customer service automation and software development. Continued monitoring of risks and development of best practices for trustworthiness, human oversight, and ethical use will be critical as this technology matures.
Sources
This article is based on reporting and publicly available information from the following sources:
Read more Artificial Intelligence stories on Goka World News.