In late 2022, ChatGPT emerged, introducing the dynamic potential of generative AI (GenAI) to the public consciousness. Now, a new frontier has opened in the world of artificial intelligence: agentic AI.
This new AI system draws on the large language models and systems that power GenAI, but crucially, it is designed to move beyond content creation. An agentic AI solution can also set objectives, action them, iterate and interact with other tools and systems as it works to achieve the user’s goal.
These capabilities have led some to predict the technology will radically change how we perform tasks and make decisions. According to Yoav Shoham, professor emeritus at Stanford University: “The potential here is real. But we need to match the ambition with thoughtful design, clear definitions, and realistic expectations. If we can do that, agents won’t just be another passing trend; they could become the backbone of how we get things done in the digital world.”
Agentic AI systems incorporate generative and extractive AI into their workflows, coordinating their use where best suited. In this post, we’ll examine the differences between the main AI solutions – agentic, generative and extractive AI — and explore the strengths and opportunities they offer. But it’s also important to note that these concepts can also work together and are not mutually exclusive.
Agentic AI is a sophisticated AI system that combines the data extraction and content creation abilities of GenAI with a reasoning engine, which simulates logical thinking and decision-making. This powerful blend enables agentic AI tools to break down complex requests (like booking a trip or writing a paper) into separate components and action each of them in turn, using the information surfaced to determine next steps. They can also connect to other systems (sometimes called “AI agents”) required to achieve the user’s goal. Its core use case include solving complex, multifaceted problems.
While agentic AI tools can respond to any query, they really excel with those requiring action or multi-step reasoning, such as open, complex or interdisciplinary queries. For example, you could enter the prompt: “Show me emerging research trends in climate science and create a comprehensive literature review with methodology recommendations.”
This varies depending on the tool’s prompt engineering (guidelines) and the nature of the user’s query. It can range from real-world progress towards the user’s goal (e.g. booking of a flight) to an in-depth report.
Generative AI is a system that generates fresh, original outputs in response to a user’s prompt. It does this by predicting what the most relevant response would be based on the knowledge and patterns it has ingested during its training. Large language models are an essential component of GenAI tools, enabling them to understand natural language requests and generate easy-to-understand responses.
Generative AI is useful when the user wants new content created. For example, you could enter the prompt: “Write a summary comparing the key findings from these 10 climate papers.”
This depends on the tool’s prompt engineering (guidelines) and the user’s requirements. While the form of the output can vary – common options include text, images, video and audio – the content generated is always new and original.
In use for more than a decade, extractive AI technology acts like a helpful librarian, quickly scanning large volumes of data to identify the most relevant phrases, sentences or data points, which it then extracts and connects. It can also generate summaries using the extracted content. It is typically used to answer questions, highlight connections and provide concise summaries.
Extractive AI is great for “find me the answer to this question” queries. For example, “What are the main conclusions in this specific research paper?”
This varies depending on the design of the tool and the user’s query. Typically, users see a subset of the original input data pulled out, highlighted or reorganized: Extractive AI never generates new.
Agentic AI represents the next step in the evolution of artificial intelligence, building on the foundations of generative and extractive AI to create systems that achieve real-world goals. For academics and librarians, this represents an enormous opportunity to streamline and improve their research activities – a topic we will touch on in future blog posts.
To discover more about how Elsevier uses these exciting technologies, please visit Elsevier.com, where we have posted a PDF to dig deeper into these different forms of AI. You can also visit our Deep Research page, which explains how we’re incorporating these technologies into our various solutions.
Shoham, Y. (July 2025). Don’t let hype about AI agents get ahead of reality. MIT Technology Review. https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/
