Episode 6: How does AI become agentic?

Hi everyone, this is another episode of Tech Break by Friday. Where does the term Agentic AI come from? What are AI agents, and why use them instead of foundation models?

Psychologist Albert Bandura introduced the concept of agency. Agency in individuals is described as the ability to act in full or partial autonomy. So, AI agents are AI systems that can perform a certain number of tasks autonomously.

How does an AI Agent System problem solve:

1. Perceive: The system gathers information from its environment. In this stage, it will try to find key elements in the data, depict categories, or determine the data's relevance.

2. Reason: An LLM acts as the operator or reasoning engine. It understands the tasks given, allowing tools to be utilized to produce results and reach the final solution. One use case would be to use RAG to determine how to solve a specific problem from the data.

3. Act: It makes decisions and takes appropriate actions. By integrating custom implemented or external tools, agents can act with them to get a result for the problem presented. If the agent receives a prompt or an instruction on how to solve the specific issue, it will act on the instructions. This removes the manual work of maybe calling different tools and collecting their results.

4. Learn: Usually, there is also a feedback loop. This loop allows the agent to improve on the tasks for next time. It could be like a flywheel since the data from the agent’s iterations is fed back to the system to enhance the agent.

(Source: Pounds, E. ,2024)

A powerful solution for businesses to drive faster decision-making and operate more efficiently.

As humans, we use our short-term and long-term memory to remember things more easily. Short-term memory helps us remember the context of a conversation with another human. In agents, short-term memory can be used to keep the conversation context in the same thread as customer bots. I want you to know that AI agents remember the information that you have provided in the conversation previously. But we also have long-term memory. A direct equivalent to long-term memory for AI agents is a vector store that contains documents transformed into vectors. The agent has access to this data store to use at all times, and they can act upon them.

AI agents can act autonomously, but they need an LLM as their main component. Their ability to reason depends on the large language model used for this specific task. Agents have a memory to act on the context provided and/or the external data. Their ability to solve problems is simulated from how we humans observe, reflect on the world, and decide if a certain action will solve the problem. In case of success, we learned how to solve the issues, and in case of failures, we knew what wouldn’t so we could try something else.

References:

1. Pounds, E. (2024, October 22). What Is Agentic AI? NVIDIA Blog. https://blogs.nvidia.com/blog/what-is-agentic-ai/

2. Rogerson, K. (2024, October 15). Agentic AI Explained: Definitions, Benefits & Examples of AI Agents. Interface.Ai. https://interface.ai/blog/what-is-agentic-ai/

3. (24) Understanding Agentic AI: Origins, Applications, and Its Near-term Impact on Human Resources | LinkedIn. (n.d.). Retrieved March 26, 2025, from https://www.linkedin.com/pulse/understanding-agentic-ai-origins-applications-its-near-term-hill-ir6fc/

4. Bandura, A. (1997). Self-Efficacy: The Exercise of Control. W.H. Freeman.

5. Skinner, B.F. (1971). Beyond Freedom and Dignity. Knopf.

6. Marcus, G. (2019). Rebooting AI: Building Artificial Intelligence We Can Trust. Pantheon.

7. Boden, M. (2018). Artificial Intelligence: A Very Short Introduction. Oxford University Press.