MIT: Professor Phillip Isola Explained What Agentic AI Really Is

Computer scientist Phillip Isola, associate professor in the Department of Electrical Engineering and Computer Science at MIT CSAIL, in an interview with MIT News broke down what the term agentic AI ("agentic AI") actually means and how the real capabilities of such systems differ from marketing hype around them.

What is agentic AI

Unlike an ordinary chatbot that simply responds to a user's message, an AI agent is capable of independently planning a sequence of actions: breaking a task into steps, calling external tools, browsing web pages or executing code — and adjusting the plan as it goes. It is this ability to act in multiple steps and autonomously, rather than simply generating text, that distinguishes agentic AI from classical large language models.

• Interview subject — Phillip Isola, associate professor at MIT CSAIL
• Interview topic — what agentic AI is today and how researchers want to see it
• Publication — MIT News, AI research section

How agents differ from hype

Over the past two years the industry has released dozens of products under the banner of "AI agent" — from browser assistants like OpenAI's Operator to computer control features from Anthropic and autonomous coding agents like Devin. But not all of them are equally autonomous: some systems merely simulate agency through hardcoded scripts, while researchers like Isola speak to a more fundamental question — how to teach a model to truly plan and adapt to new situations, rather than simply execute a preset algorithm.

Isola himself is a specialist in computer vision and generative models, known among other things for work on image-to-image translation algorithms. Such background in fundamental research of generative systems gives him grounds to view the agent craze with detachment — not as a product marketer, but as a researcher for whom the boundary between "model responds" and "model acts autonomously" is a matter of rigorous technical analysis, not advertising slogan.

Where the field is heading

The key question posed by the MIT News headline is not just "what is agentic AI today," but "how we want to see it." This underscores that the development of agent systems is not only an engineering task but a normative one: the research community still needs to determine what level of autonomy is safe and useful before such systems receive greater independence in real tasks — from booking tickets to managing finances or code in production.

What this means

While the market is flooded with products labeled as "agent," the sober view of MIT scientists reminds us: true AI autonomy is an open research question, not a solved technology that can simply be packaged in an interface and sold.