There has been a recent explosion in the deployment of automated software systems called AI agents. A report from November 2025 MIT Sloan School of Management and Boston Consulting Group found that 35 percent of businesses surveyed had already deployed AI agents, while another 44 percent planned to implement agentic AI soon.
Understanding the fundamentals and potential impacts of these increasingly popular devicesMIT News We spoke to Philip Isola, an associate professor in the Department of Electrical Engineering and Computer Science (EECS) and member of the Computer Science and Artificial Intelligence Laboratory (CSAIL), who studies the intelligence possessed by AI agents as well as the underlying models and mechanisms that power agent AI systems.
Why: What is Agentic AI and how is it different from Generative AI models like ChatGPT and Cloud?
A: Agent AI is AI that takes action in the world. These actions can be physical actions, such as robotic manipulation, or digital actions, such as booking a flight. On the other hand, we think of generic AI as creating stories, poems, art, and pictures rather than taking actions for us.
The term “Agent” is just a brand name. Usually this means AI that will help people interact with an application, website, or the physical world. Most of the agents we come across today are digital agents, like customer service agents you can talk to about product complaints.
Most companies offering agents use some of the same AI models under the hood and give them the ability to take actions and remember what happened. An agent basically starts with a fundamental generative AI system, like a cloud. Companies then put different wrappers around that foundation model for their product or application. Those wrappers can be specific tools that the agent can use, and those tools depend on the application. The agent may have access to a calculator so that he can solve math problems, or he may have access to more complex hard drives and operating systems so that he can remember a firm’s financial data and past business interactions.
The biggest challenge in developing agentic AI comes from the lack of training data. If I wanted to create a system that could go online and book a flight for me, that sounds pretty simple. But we don’t have much data that tells us how to do this – where to move the mouse, which buttons to click, what to do if something goes wrong, or how to call someone and negotiate the price of an airline ticket. One way to train such a system is to have AI agents visit airline websites, try things out, and see what works and what doesn’t work. These environments are difficult to model, so often the agent must learn by trial and error.
Why: What are some promising applications of agentic AI?
A: I think the area where we’ve seen the most success is with coding agents. This is something that has evolved from Generative AI. People trained language models on code, and then they could predict what a human would do to solve a coding problem. Furthermore, an agent can learn to do this by going through a feedback loop where it tries different solutions and checks to see if it got the answer right. As long as it can check the answer, the AI agent can execute this trial-and-error loop until it figures out a good strategy.
But there’s always a balance between automating decision making versus simply assisting and informing humans. Analytical AI methods, like systems that help predict possible outcomes of decisions, are not agentic in nature, but are very informative to human decision makers. For cases that are either high-risk or safety-critical, such as medicine, security, high-level trade policies, etc., the technology may not be ready for AI to fully automate those processes, or we may not even be comfortable with it.
Why: Should we think about such risks when using AI agents?
A: A big risk area comes from the fact that it is often very easy to get agents to do certain types of work for you. With coding agents, you can “vibe code” and ask the agent to create a code for you, so you don’t have to do the hard work yourself. A bigger risk is that, because it’s so simple, people won’t put enough effort into verifying that it’s working right. Bugs will be introduced, private data will be leaked – it’s already happening.
Agents are not perfect, in the sense that they can make mistakes because they are not well trained and do not know what to do. But even if they are very capable, if a human does not use them appropriately or gives them instructions that are too vague, the AI agent may make a mistake because the human made a mistake. If humans were less involved in thinking about all the consequences, I think we might be more prone to making those mistakes.
An additional factor is the risk of de-skilling. It’s not clear how far this will go, but as we rely on agents to do our homework, our coding, and our math, we may lose the ability to do that ourselves, and we may lose that ability very soon because technology is not yet ready to fully automate those processes.
Why: What is the future of agentic AI?
A: What we now think of as agentic AI refers to large language models that use tools to interact with digital and physical systems. One obvious limitation is that, under the hood, these have the architecture of a language model and are trained on text data. To create even more powerful AI agents, we may need to model video, physical forces, time series, radar scans, and other modalities. We may need models with fundamentally different architectures that can handle continuous data, high-dimensional data, stochastic data, etc.
But, on the other hand, perhaps an extremely good coding model could act as a puppet to interface with sensors, actuators, and web APIs? Perhaps, once you have a super-smart reasoning system that understands math, language, and code, you can give it a camera and a keyboard and it will figure out what to do in the spatial domain. Is the next wave of AI just going to be cloud with sensors, actuators, and tools, or is it going to be something built from the ground up in a new way? This is a big question that many people in AI are struggling with right now.
