AI Agents and Agentic AI: The Next Evolution of AI

In our modern world of AI, “AI Agents are software systems that can process information in a given context and act on behalf of the user”. This is a very condensed variant of the definition, so let’s unfold it.

• AI Agents are software systems: The most well known AI Agents are the ones we know as ChatGPT, Gemini, Claude and so on. These are all accessed through a digital interface like a web browser.
• Can process information: This basically means that AI Agents consume input from for instance the users. We often know this input as our prompt or our questions to the AI, but it can also be large spreadsheets of data, video recordings, entire written documents and so on. In other cases the parts of the information might be predefined, and act as base information. The magic is when the processing adds the large language models (LLMs). This means that the information given to the AI Agent is being ‘understood’.
• Context: Simply put, this is the short term memory of the AI Agent. When the AI Agent is provided information, like an instruction from a user, it will review its ‘memory’ to give the user the most relevant response. If you use the same AI Agent like ChatGPT or Gemini over and over, you might have realized how it’s referencing past conversations or topics you have asked it about.
• Act on behalf of the user: Depending on what the AI Agent is exactly configured to do, it will use the components covered above to best respond to the prompt of the user. Sometimes it feels like the AI Agent is acting on its own behalf, but that is likely the result of the reasoning it’s been through and how it ‘thinks’ it’s helping the user the best.

There is no better way to wrap this up, than to use an example.Answer My Emails is an AI Agent which will respond to your emails. AI Agents doing this actually exist in different shapes and versions. This one is made up.

The Agent is adding the content of the emails the user is receiving to its context (short term memory), as well as the metadata of the emails, like when they were received and who the senders are. This is now being processed together with the instructions (information), which could be predefined for this Agent. The instructions could be something like:

“As my personal assistant with access to my email box, please respond to my incoming emails in an informal and down to earth way and request the sender for more information if needed to respond to the enquiry. For meeting requests, check my calendar and suggest this time to the sender.”

While this is a fairly simple instruction, it should serve the purpose of the example. For the Agent to be able to act on the behalf of the user, it must be able to access the email. This is done as a part of the setup - like when the user signs into the mailbox with a user id and password.

While the Agent will now answer emails the user can always interfere and pause the Agent and take part in answering the emails.

While this is describing one simple Agent, this next generation of autonomous can not be underestimated. This serves as a natural segway into Agentic AI.

If the introduction to AI Agents above did well, it should be no challenge to understand the concept of Agentic AI. Where AI Agents have a somewhat narrow scope of what they can do, Agentic AI is about more complex autonomous reasoning, planning and decisions making for instance by interconnected Agents that can interact with each other, to solve more complex tasks with larger scopes. See AI Agents as modules or bricks, and put together they form the concept of Agentic AI.

While it is hard to point down an industry standard for defining Agentic AI, the concept can become a little more clear and even intuitive to understand when broken down, by another example.

Let’s imagine an Agentic Inventory Management System. This interconnected system handles all the steps related to keeping the inventory up to date and breaking it into subtasks basically starts to define the agents. The system must handle:

• Check inventory status
• Update inventory status
• Order products
• Return products from customer
• Forecast demand
• Get product sales data

These subtasks will do for our example.

While each of these individual tasks could be separated into AI Agents that humans could interact with, they could also take part in an interconnected Agentic system. The system would analyse the sales data, plan restocking, decide when to order new goods and execute the order, update the status of the inventory when items are either out of stock, on its way or in stock and so on.
With the Agentic system being instructed to keep the inventory in a healthy state which supports the business operation, the system would act autonomously with the tools (the agents) made available to it.

A bit like a chain reaction, one task would be solved, and the output of checking the status of product A for instance, would maybe lead the Agentic system to update the planned ordering schedule or maybe check the sales data to forecast when the product is expected to be out of stock again.

Over time the system could benefit from connections to the CRM system (or Agentic CRM system, of course) and the financial reporting data and this way give the Agentic system more context to help the planning, reasoning and decision making.

While everything AI can seem a bit overwhelming because of the technology itself and the speed its developing, getting started is not as difficult. Putting a side the challenges of organizational structures, aligning ownership of overlapping topics and domains and other cooperate blockers, any business in any industry can started by:

• Mapping tasks and scoring them by how repetitive they are, the scale of the task and how difficult the tasks are. The first AI Agents are likely for very repetitive tasks that’s done often
• Define dependencies to successfully execute those tasks. If it’s data, a system or some input. This will help understand what an AI Agent must deal with and potentially integrate to
• Lastly, decide on implementation. The community and industry is booming with options. By having business requirements prepared upfront it’s easier to have conversations with internal development resources and external implementation and tech partners.

Don’t try to start with a full blown Agentic AI system, but start small and get experience with the technology. This way it’s also easier to discover what requirements the business does not meet, and defining the specific needs are a more tangible task when working with real use cases.

Generative AI will without doubt change a lot of things. One of the things being how businesses operate and interact with their customers.
Understanding the deep details and technical pieces of AI Agents and Agentic AI can be overwhelming for some. But like with other revolutionary technologies such as the Internet, understanding and starting to adopt will be important to keep up with competition and expectations from the customers. Fugentic makes it easy for any business to adopt and run AI solutions, so they can stay up to date and keep serving their customers effectively. The platform allows businesses to connect their business services to the AI ecosystem, so customers and prospects can continue to find business and brands when they take their research and browsing to AI. Sign up to the free BETA platform here.