I was searching for a way to explain our AI agents to folks while also communicating that those agents are only one part of the puzzle, I found a recent research paper from Cornell University titled AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications and Challenges by Ranjan Sapkota and colleagues, that I think is helpful.
While helpful, this paper isn’t light reading—it’s dense, academic, and packed with technical nuance, so in this article I try to extract the basics, because beneath the jargon lies a very useful framework for understanding how AI is evolving. Let me try to break it down in a way that’s practical and hopefully simple enough to allow you to sift through the hype as you start to see more and more coverage of this new tech.
AI Agents: Your Trusty Personal Assistant
Imagine you hire a super-efficient personal assistant. This assistant is great at specific tasks—scheduling meetings, answering customer emails, or pulling data from your CRM. They follow your instructions to the letter, work independently once you’ve set them up, and adapt slightly if a meeting time changes or an email needs a different tone. This is essentially what an AI Agent is in the world of artificial intelligence.
According to the paper, AI Agents are modular, task-specific systems built on top of large language models (LLMs) like GPT-4. They’re designed to handle well-defined jobs with a degree of autonomy, reactivity, and focus. At NewTide we think of them as digital workers who excel at one thing at a time. For example, an AI Agent might power a customer support chatbot that answers FAQs, queries a database for order statuses, and escalates complex issues to a human—all without needing constant supervision.
What makes AI Agents special is their ability to act like a bridge between raw AI power (like LLMs) and practical, everyday tasks. They’re not trying to boil the ocean—they’re focused, efficient, and built to make your life easier in specific ways. In my work with clients we are seeing AI Agents shine in things like automating repetitive workflows, think email prioritization or report summarization. They’re the workhorses of today’s AI landscape, and their simplicity is their strength. The more narrow the task or the ask the better the results.
But here’s the catch: AI Agents are solo performers. They’re great at their assigned tasks, but they don’t naturally collaborate with other agents or tackle complex, multi-step problems that require teamwork. That’s where the paper introduces Agentic AI. Sounds like the same thing, but the difference is important as we think about how to get much more work out of these systems.
Agentic AI: The Dream Team of Digital Workers
Picture that same helpful agent we just talked about but instead of one assistant, you’ve got a whole team—a store manager, a data analyst, a category manager, and an operations coordinator. Each has their own expertise, but they work together, divvying up tasks, sharing insights, and adjusting plans on the fly to hit a big goal, like introducing a new product to your stores. This is the essence of Agentic AI: a system of multiple, specialized AI Agents that collaborate, communicate, and coordinate to solve complex problems.
The paper defines Agentic AI as multi-agent systems where each agent has a specific role, and they work together under a framework that orchestrates their efforts. It’s like a digital orchestra, with a conductor (or “meta-agent”) ensuring everyone plays in harmony. For example, in a robotics swarm, one agent might handle navigation, another processes sensor data, and a third plans the mission—all while sharing updates to avoid collisions and achieve the goal. The paper highlights real-world applications like autonomous logistics, medical decision support, or even research automation, where Agentic AI systems break down big problems into manageable chunks and tackle them as a team.
This collaborative approach is a game-changer. Unlike a single AI Agent, which might struggle with a task requiring diverse skills, Agentic AI thrives on complexity. It’s like the difference between hiring one person to build a house (good luck!) versus a crew of carpenters, electricians, and plumbers working together. This is what has me so excited about Agentic AI when they see how it can streamline workflows that span departments—think fuel supply chain optimization or cross-functional project management.
The Evolution: From Solo Acts to Teamwork
The paper traces this shift from AI Agents to Agentic AI as a natural evolution, sparked by the rise of generative AI (like ChatGPT) in late 2022. Early generative models were like talented but passive artists—they’d create amazing content when prompted but wouldn’t take initiative or work with others. AI Agents built on this foundation, adding autonomy and tool-using skills to handle specific tasks. Agentic AI takes it further, creating ecosystems where agents collaborate, learn from each other, and adapt dynamically.
This progression feels personal to me because it mirrors how we grow from individual contributors early on in our careers, to later become managers helping guide and coordinate the work of others. As you and your business scales, you build a team, delegate tasks, and create systems to hopefully keep everyone aligned. This is the promise of Agentic AI in a nutshell: scaling intelligence through collaboration.
The Challenges: Keeping the Team in Check
Of course, no dream team is perfect, and the paper doesn’t shy away from the challenges. AI Agents can suffer from “hallucinations” (making stuff up), shallow reasoning, or brittleness (failing if instructions aren’t just right). It’s like an assistant who misinterprets your request or gets stuck on a tricky task. Agentic AI, meanwhile, faces bigger headaches: misaligned agents, unpredictable group behavior, or errors that cascade across the team. Imagine a project where one team member’s mistake throws everyone else off—that’s the kind of risk Agentic AI introduces.
The paper suggests solutions like better memory systems, causal reasoning (to understand cause-and-effect), and governance frameworks to keep things safe and ethical. For me, this resonates with the real-world challenge we always face with managing adoption of new tech. Clients often worry about AI making mistakes or acting in ways they can’t control, for now we need to keep humans in the loop and in control, with those checks in place we can have systems that are not just powerful but trustworthy.
Why This Matters
So, what does this mean for our businesses in our industry? If you’re looking to automate specific tasks—like customer support, data retrieval, or scheduling—AI Agents are your go-to. We can deploy our platform today to start building the intelligence and experience with your new digital employees right away. Just like new hires the quicker we get them onboarded and trained up the faster we can get them on the path to taking more independent actions later. These solutions are affordable, focused, and ready to deploy now. The money they save can immediately pay for getting the AI platform. By next year we will be focused on coordinating a whole fleet of agents that are already trained up on your business—Agentic AI. It’s more complex and still evolving, but it’s where the industry is headed.
Personally, I’m excited about Agentic AI because it feels like the next step in making AI truly transformative. It’s not just about replacing tasks; it’s about building systems that think and work like human teams, but with the speed and scale of machines.
The Road Ahead
The paper closes with a vision of where this is all going: AI Agents will get smarter, learning proactively and reasoning causally, while Agentic AI will scale into domain-specific systems for fields like healthcare or law, or in our case for energy, fuels and c-stores. It’s a future where AI doesn’t just assist but collaborates, plans, and even anticipates your needs.
For now, my advice is to start small with AI Agents—get comfortable with their capabilities—and keep an eye on Agentic AI as it matures. The paper’s taxonomy gives us a map to navigate this shift, and I’m betting it’ll guide us toward a world where AI isn’t just a tool but a partner.
This article is based on insights from “AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications and Challenges” by Ranjan Sapkota et al.
