Understanding Agentic AI: The Evolution from Chatbots to Autonomous Digital Assistants

For most individuals, initial interactions with AI agents are likely to occur in areas such as online shopping. (Representational)

Technology enthusiasts searching for the next innovation frontier are increasingly turning to "agentic AI" as the future. This term dominates marketing campaigns and tech industry apparel.

The definition of what makes an AI product "agentic" varies depending on the vendor, but generally promises advancement beyond today's generative AI chatbots.

While chatbots provide valuable services, they remain limited to communication. They can answer queries, retrieve and condense information, generate written content, images, music, videos, and code. AI agents, however, are designed to independently execute actions on behalf of users.

This distinction has created widespread confusion. Google searches for "agentic" have surged from virtual nonexistence a year ago to peak interest this fall. Merriam-Webster hasn't formally added the term to its dictionary but identifies "agentic" as emerging slang defined as: "Able to accomplish results with autonomy, used especially in reference to artificial intelligence."

A recent report released by Massachusetts Institute of Technology and Boston Consulting Group researchers, based on surveys with over 2,000 global business executives, characterizes agentic AI as a "new class of systems" that "can plan, act, and learn on their own."

The MIT Sloan Management Review report states: "They are not just tools to be operated or assistants waiting for instructions. Increasingly, they behave like autonomous teammates, capable of executing multistep processes and adapting as they go."

AI chatbots—such as the original ChatGPT launched three years ago—function using large language models that predict subsequent words based on extensive training data from human writing. Though remarkably human-like, especially with voice capabilities, they essentially perform sophisticated word completion.

This differs fundamentally from what AI developers—including OpenAI, Amazon, Google, IBM, Microsoft, and Salesforce—envision for AI agents.

"A generative AI-based chatbot will say, 'Here are the great ideas' … and then be done," explained Swami Sivasubramanian, vice president of Agentic AI at Amazon Web Services, in a recent interview. "It's useful, but what makes things agentic is that it goes beyond what a chatbot does."

Sivasubramanian, a longtime Amazon employee who began leading AI agent development in Amazon's cloud computing division earlier this year, sees tremendous potential in AI systems capable of receiving a "high-level goal," decomposing it into sequential steps, and executing them. "I truly believe agentic AI is going to be one of the biggest transformations since the beginning of the cloud," he stated.

At its core, an AI agent functions similarly to traditional computer programs that perform specific tasks. However, when combined with large language models, it can search for knowledge enabling task completion without explicit step-by-step instructions. This means rather than merely helping draft an email, it can theoretically manage the entire process—receiving messages, determining appropriate responses, and sending replies autonomously.

For average consumers, initial experiences with AI agents will likely involve online shopping scenarios. Users could set budgets and preferences, allowing AI agents to make purchases or arrange travel using payment information. Eventually, these agents may handle more sophisticated tasks with access to personal computing systems and defined parameters.

"I'd love an agent that just looked at all my medical bills and explanations of benefits and figured out how to pay them," or another functioning as a "personal shield" against email spam and phishing attempts, said Thomas Dietterich, professor emeritus at Oregon State University with decades of experience developing AI assistants.

Dietterich questions companies using "agentic" to describe "any action a computer might do, including just looking things up on the web," but remains enthusiastic about AI systems with the "freedom and responsibility" to refine objectives and adapt to changing conditions while working for users. These systems could potentially coordinate "subagents."

"We can imagine a world in which there are thousands or millions of agents operating and they can form coalitions," Dietterich noted. "Can they form cartels? Would there be law enforcement (AI) agents?"

Milind Tambe has researched collaborative AI agents for three decades, since the inaugural International Conference on Multi-Agent Systems in San Francisco in 1995. Tambe finds the sudden popularity of "agentic" as an adjective "amusing." Previously, this agency-describing term was primarily found in academic disciplines like psychology or chemistry.

Computer scientists have debated the definition of an agent throughout Tambe's research career.

In the 1990s, "people agreed that some software appeared more like an agent, and some felt less like an agent, and there was not a perfect dividing line," said Tambe, a Harvard University professor. "Nonetheless, it seemed useful to use the word 'agent' to describe software or robotic entities acting autonomously in an environment, sensing the environment, reacting to it, planning, thinking."

Prominent AI researcher Andrew Ng, co-founder of online learning platform Coursera, advocated popularizing the adjective "agentic" over a year ago to encompass a broader range of AI capabilities. At that time, he appreciated that mainly "technical people" were using the term.

"When I see an article that talks about 'agentic' workflows, I'm more likely to read it, since it's less likely to be marketing fluff and more likely to have been written by someone who understands the technology," Ng wrote in a June 2024 blog post.

Ng did not respond to inquiries regarding whether this view still holds.