The Agents are Coming!

The agents are coming, and it is an invasion. The way that humans interact with AI and the way that AI interacts with humans shifted radically with the introduction of generative AI. After two full years of high speed acceleration, a natural language interface and growing generative capabilities make the new technology an onramp for the rapid shift to AI agents. Just as an agent acts on my behalf, with full awareness of my capabilities, history, and the potential value I can bring to a client, AI agents will eventually represent people, departments, organizations, and ecosystems with intelligence far beyond what we can amass and retain ourselves.

In all my conversations about agentic AI, I have yet to find anyone who disagrees with the fact that one day there will be more AI agents employed by corporations than there will be human employees. We are already seeing a blitz of task-based agents to do everything from write code, develop applications, write books, make travel arrangements, or write a business plan and strategy. As task agents are bundled together, we are also beginning to see the introduction of persona agents, agents that mimic more complex tasks performed by humans and that fill the roles played by individuals in an organization.

The AI Driver Agent

The best example of an AI agent is what most people call a self-driving car. Companies like Waymo and Tesla are already making autonomous cars a reality, but the cars don’t drive themselves. These companies have created AI agents that mimic the myriad of actions taken by the driver of a car. In order to drive, the agent needs to be able to perceive the world around the car and constantly react to what appears. The agent needs to be able to steer, accelerate, and brake with smooth precision. It needs perception, context, and the ability to act and react based on thousands or millions of possible situations and outcomes, something the human mind does almost automatically. To train these AI drive agents, it was necessary to simulate multiple possible worlds and do extensive scenario planning. We call these new vehicles “self-driving cars,” but in reality, they are cars that are being driven by an AI agent.

What is an AI agent?

Because AI agents are the latest craze following generative AI, there is some confusion in the market. For example, some vendors with generative AI solutions are now calling their applications agents. While there is room for a technical and theoretical discussion, it is best to define agentic AI conceptually, focusing on the concept of agency. Words matter.

With this approach in mind, an agent is an intelligent software program that mimics the way humans accomplish tasks, fill roles, work as departments, and collaborate in complex ecosystems. Therefore, an AI agent must be intelligent, autonomous, active, reactive, interactive, adaptive, and limited.

Intelligent – an agent possesses knowledge, is able to reason, especially for making decisions, and can acquire new knowledge, either through communication with data services, with other agents, or entirely on its own.

Autonomous – even though the agent’s objectives and operating procedures are defined in advance, once set in motion, an agent operates on its own, independent of human intervention.

Active – an agent must be able to act, to perform tasks that impact the environment in which it lives.

Reactive – an agent must be able to react to changes, events that are taking place in the environment in which it lives, or other agents.

Interactive – an agent must be able to communicate with other agents, sending and receiving understandable messages, and being able to act based on the interaction.

Adaptive – an agent must be able to adapt to the environment in which it lives based on inputs from humans, other agents, and data services.

Limited – an agent must have limitations to what it can do. They operate based on a clear definition of what an agent can do and what the agent cannot do. Limitations must be set to allow the agent to operate freely and understandably in a community of agents.

The Types of AI Agents

There are four different types of AI agents: task agents, persona agents, workgroup agents, and ecosystem agents.

Task Agents

Task agents use artificial intelligence to mimic tasks normally performed by humans, applications, or devices. They are the evolution of robotic process automation, and they typically emerge wherever there is a rich set of data already curated and available for the training of the AI agents.

For example, we are already seeing agents emerge to do simple tasks like booking a flight, renting a car, checking grammar, summarizing documents, sending email, planning meals, organizing shopping, and drawing diagrams.

Task agents were the first to hit the market, since they are designed to mimic very specific and definable tasks where data is readily available to train models. We will continue to see an explosion of task-based agents throughout 2025 and beyond. We are already seeing vendors, individual software engineers, and data scientists stitching together multiple task agents to take more complex actions.

Persona Agents

Persona agents use artificial intelligence to mimic tasks performed by humans in a specific role within an organization, or as part of their everyday life. They are typically built on agentic architectures created to combine a series of tasks, typically related to a single area of operation.

For example, we are already seeing persona-based agents for roles like a business analyst or software engineer. A business analyst agent does all of the work you would expect from a human business analyst. They are able to use business intelligence and analytics tools to find answers to any question that a data or business person might ask. The more advanced business analyst agents allow users to ask questions without having to phrase their questions to fit specific SQL-like language and without prior knowledge of the data or the business. They are designed for true human interaction.

In like manner, a software engineer agent does all the work of a software engineer. In the same way that a software engineer receives a set of product specifications from product owners and architects, these stakeholders can pass their requirements on to a software engineer agent, and the agent will write working code. The more advanced software engineer agents can write code based on input that is given from the perspective of the business, not necessarily including all the jargon typically used by more tech savvy users.

Persona agents have been maturing quickly over the last couple of years and will continue to mature in 2025. This type of role-based agent has also been limited to areas where there are rich sets of data and highly digital output. In 2025, based on advances in LLM “chain-of-thought” capabilities, we will see persona agents extend to even more roles, including those roles with far less digital output.

Workgroup Agents

Workgroup agents knit together a community of persona agents and task agents to mimic the work done by an entire workgroup, department, or community of people.

For example, we are already seeing multiple persona agents emerging in the marketing and sales organizations. In the example of marketing, agents are emerging for content creation, campaign orchestration, advertising, and sales qualification. It won’t be long before these different personas are integrated with the ability to collaborate for end to end marketing. While this analogy does not represent the work of the entire marketing department, it is an example of how workgroup agents will eventually morph into more complex execution.

We are still early in the release of workgroup agents, but in 2025 we will see progress from two different directions. Application vendors are already working to move from automation in their platforms to more agent-based execution. We are also seeing the emergence of an ever increasing number of new, agent-first software vendors entering the market, seeking to replace traditional applications with agentic architectures at their core.

Ecosystem Agents

Ecosystem agents knit together an entire network of workgroup, persona, and task agents to simulate the work done by an extended group of organizations or communities.

For example, new vendors are already emerging to knit together software engineering. The entire software engineering lifecycle (SELC) includes business sponsors, product owners, architects, software engineers, infrastructure engineers, and operations engineers. The SELC represents a complex ecosystem that requires visibility and intelligence across the entire ecosystem in order to streamline time to delivery and deliver high quality products. By understanding the most important roles and the most highly leverageable task points in the SELC, these new vendors are already building entire agentic ecosystems to support agents for product ownership, architecture, engineering, and operations.

While there are some application companies attempting to build out agentic ecosystems, the early versions are limited to highly digital roles and departments. By the middle of 2025, emerging software vendors will come out of stealth with agent-first ecosystem agents, where agency is undertaken across entire business lifecycles. In order to advance the spread of ecosystem agents, agentic architectures must mature to include specialized orchestration capabilities like agent communication protocols, agent observability, governance, and auditability, along with very detailed human in the loop planning and feedback.

The Best is Yet to Come

We are still early in the productization of AI agents. However the speed at which AI technology is advancing mandates action for executives, product leaders, and analytics leaders in all industries. Ferraro Consulting recommends that you begin with education. AI Literacy is of the utmost importance in 2025. Along with learning, leaders can begin experimenting with generative AI as a precursor to agentic AI. Familiarize yourself with the possibilities for productivity and creativity gains possible with broadbased use; and roll out practical education for your workforce. In addition, keep your eyes out for AI agents as they begin to emerge. They will arrive in exactly the order already described: task agents, persona agents, workgroup agents, and ecosystem agents. Look especially for agentic AI with application directly to your industry, or your area of the organization. Just as we have seen the move from large language models with broad intelligence to small language models with specific domain expertise, we will see the same type of rollout for agentic AI.

For more information on Ferraro Consulting, send a message to info@ferraroconsulting.com.

For more information on our senior level course, AI Literacy for Executives, request a meeting at TheNewLiteracy.com.