A Framework to Make the Complicated World of AI Agents…Less Complicated

The Premise

AI Agents are creating extreme efficiencies in certain tasks and are set to grow in their capabilities. According to an April 2025 Markets and Markets report, AI Agents Market, the AI Agents market is projected to grow from USD 7.84 billion in 2025 to USD 52.62 billion by 2030. This notable expansion, characterized by a robust CAGR of 46.3% between 2025 and 2030, is mostly driven by advancements in foundational models.

There are many definitions of AI Agents, so to be clear in this article I assume an AI Agent is an autonomous, intelligent software system designed to perform specific tasks or make decisions without continuous human intervention. Powered by artificial intelligence, including machine learning, natural language processing, and reasoning capabilities, AI Agents interact with their environment, process data, and execute actions to achieve predefined goals. Examples include virtual assistants, customer service chatbots, recommendation systems, and autonomous workflow managers. AI Agents are characterized by their ability to perceive, reason, learn, and adapt, enabling them to handle complex, dynamic tasks with increasing efficiency and accuracy.

We know there are many AI Agents available for license, to which business leaders need access to maintain market competitiveness. One can see the Cambrian explosion of new AI Agents and AI Tools popping up all over the place. From AI Sales Agents to Coding and Research, there are literally hundreds of categories. One of my favorite lists of the AI Agents Market Landscape, is the AI Agents Directory.

What does this mean for business leaders? How can we put all this into a framework by which business leaders can understand and make decisions around capital allocation, talent needs, investment decisions, product roadmaps…etc.?

Models vs Agents and the Agentic Maturity Model

In the first quarter of 2025, the market saw GenAI Models being enhanced with Tools and Orchestration layers in order to become AI Agents, capable of independent value creation outcomes. AI Agents are not just smarter GenAI models, they are a completely different evolutionary value creation entity. They use GenAI models combined with access to tools (like Travel & Banking Services) as well as an orchestration layer of software (workflows) to act on behalf of the user, to autonomously deliver valuable output. As a simple example, an AI Agent can actually book flights for someone, versus an AI Assistant which would just tell them which flights they should book, based on input from the user and the trained model’s capabilities to recommend. Furthermore AI Agents are not just hard coded solutions, AI Agents are nondeterministic; they deliver original content custom to each interaction even if the input from the customer/user is the same.

Today, with the advent of MCP (Model Context Protocol) Servers, the ecosystem of tools accessible by Agents is growing rapidly. This is amplifying the AI Agent ecosystem development by allowing the AI Agents to plug into legacy systems (Banks, Travel Systems, CRMs...etc.) to become more effective and autonomous at creating the desired outcomes it is designed to produce.

To add even more acceleration, the market is now trying to coalesce around a common method of Agent-to-Agent interaction, which will again exponentially increase the compounding power of an Agentic Workforce by allowing a common way for Agents to interact with each other. The beginnings of this can be seen with the release of A2A by Google and other recent Agent-to-Agent frameworks. As organizations recognize the value of AI agents, they integrate multiple agents into cohesive systems. These agents communicate, share data, and collaborate to tackle more complex workflows, such as supply chain optimization or multi-step project management.

This gets very confusing very fast. How can we think about AI agents in a way that allows business leaders to make proper decisions? In an effort to align with the long proven business idiom – “Simple isn’t easy”,  I like to think about AI agents’ growth path similar to the maturity of an organization of people, so I have created a directionally correct, but grossly simplified model for thinking about how Agents will progress over the next few years. I like this method because it anchors on a known business concept (talent allocation and investment) to guide the development, investment profile and use of AI Agents by business leadership.

I call it The Agentic Maturity Model, and it tracks the growth of AI Agents over time and maturity, segmenting the technology based mostly on output capabilities.  It starts with GenAI Models, which consist of a database and an LLM or other AI powered model. Add a user interface (UI) of some kind (chat or voice) and you have an AI Assistant. Add to that, a software orchestration layer and MCP access to various Tools (like financial platforms or travel platforms), and you have an AI Agent. At this point we have moved beyond a traditional digital tool in terms of outcome generating capabilities. An AI Agent is more akin to a person in the outcomes and tasks they can perform, than a legacy digital tool.  If we ask those AI Agents to interact with each other to solve a common problem, we have a Multi Agent System (MAS). This is similar to the time-tested, human based pizza fed “Tiger” team, tackling a set of tasks but bringing different viewpoints and assets to the table based on their foundational model training. As an example – you could have a MAS creating marketing content with team of Marketing Research, Design, Image, Video and RAG Agents all working collaboratively. Finally, when we add a hierarchy of MASs with supervisory Agents ensuring compliance and outcomes that meet the intended purpose…, we have an Agentic Workforce.

AI Agents are New Value Creation Assets in the Workforce

This simplified taxonomy should make it clear there is a difference between GenAI Models, AI Assistants and AI Agents. That the differences are both a technical, architectural difference, and a functional difference that business leaders must recognize require thinking about AI agents differently that just iteratively better GenAI Models/AI Assistants or legacy digital tools. When one thinks about AI Agents, they seem to be sitting somewhere between traditional digital tools and humans when it comes to ROI and value creation. AI Agents far surpass humans in cost efficiency, but are not nearly as capable, while simultaneously being well beyond traditional digital tools in effectiveness and flexibility because they can produce actual output and navigate uncertainty. This is why AI Agents should be treated as a new type of value creation asset for the business world and should be invested in with a new investment profile. Very similar to how budget for talent is treated separately from working capital for digital tools.

For business leaders this means we have new aspects to consider when allocating working capital in our organizations with an Agentic Workforce. To name just a few:

1. Return on Investment - Businesses need to think about how to properly measure the ROI of an AI Agent. Is it cost based valuation or outcome-based valuation? If it means my sales reps can now handle 30 customers vs. 10 customers per week, yet it means their email and phone interactions can be better researched and better delight the customer, I cannot choose only cost based valuation as the AI Agent has not just added more client interactions, but simultaneously enriched those client interactions.

2. Trade Offs – The market is flush with innovation. When is it the right time to build vs buy? Who makes that decision in your organization? Is it the functional leader, like the CMO who needs content creation AI Agents for her business or the CIO, whose internal AI Tool development team has the ability and current budget to develop such a content creation AI Agent?

3. Function Prioritization – What functions in my organization can and should be moved over to AI Agents, and how? AI Agent capabilities are advancing so quickly, how can I predict where things will be in six months?

4. Organizational Budget Structure – If less talent is needed to scale certain functions, and that working capital is now shifted to AI Agent development which is centralized in the CIO’s office, is the internal budgeting system for my organization set up to ensure AI agent funds are maintained by the functional leaders who need to control them? Do leaders need to rethink how budget for AI Agents is allocated?

5. Newfound Liquidity & Cost Controls – If an AI agent’s use can be tracked down to the token level (which it can), top business leaders in the organization have more visibility and control over how much they spend on the tasks AI Agents perform. Much more than if they were done by humans. If AI Agents grow to a significant portion of the budget, the ability to manage working capital month by month from the top of the organization becomes much easier. The liquidity of working capital that would have been locked up in less variable costs, like talent, is now more liquid. If one needed to shave 2% off of an organization’s costs for a single quarter, now they can do so by limiting AI Agent usage. One can literally turn on and off or scale down (or up) spending that was previously a semi-fixed cost attributed to talent.

6. Workforce Enablement – How are corporations going to retrain their existing and future talent to engage with AI Agents? How much effort should be put into this need and when?

 I could go on for quite a while, but I will spare you…for now.

The Timeline

From roughly 2022-2024, AI Assistants were the focus. These GenAI Models combined with a UI are extremely useful, and when engaged through a chat or voice user interface, they are certainly very effective as standalone AI Assistants. The market has not fully recognized the value of these AI Assistants yet, and there is still massive value to be explored in deploying them. As an example, an AI Assistant model accessed through chat saves immense time and costs when deployed by a Human Resources (HR) department for answering questions about the employee handbook, or as a L1 support solution for customers inquiring about a mature, well documented product line. However, very quickly these AI Assistants will migrate naturally to an AI Agent format, which will be more flexible and capable. For example, in our HR use case above, if the AI Assistant is upgraded to an AI Agent, it should be able to take action on behalf of the employee, registering for parental leave, or making changes to 401k beneficiaries…etc. not simply answering queries. The Customer Support AI Agent mentioned above may be able to resolve complex problems for users, leveraging available tools for customer remediation, not simply supplying relevant information for the customer to troubleshoot.

This new era is upon us, the epoch of AI Agents. 2025+ are the years of Agents, and it is my opinion that most corporate home-grown AI Agents deployed today are expensive science projects yet to reach productized scale. We have seen this with numerous AI agents in the news, many of which are exciting but require detailed, custom engineering to deploy the proper safeguards required to maintain privacy, security and tool interoperability. Smaller, more nimble AI Agent vendors are popping up to fill this gap, but it is the wild west right now. What is needed is an industry-wide solution for turning AI Agents from expensive science projects into scalable, value output generators that intake the relevant, enriched data from customers, are properly secured, governed, and met with a human talent base that knows how to use them. At that point, the hype of AI Agents will meet up with the value output promised.

This Agentic transition will take time. A lot longer than many people think, and the market hype will come much faster than the mainstream commercial products will.  Below you will find the general technology maturity pattern and market sentiment I believe will occur around the development of AI Agents. Though the specific time intervals will certainly not be precise, I believe the market must go through these phases of Agentic Development.

A Personal Plug - Visibility for AI Agent Governance is Key

Moving out of the trough of disillusionment, and making enterprise grade AI Agents capable of monumental costs saving and new customer value creation requires many critical needs the market is currently working on, like MCPs, ACPs and Agent-to-Agent frameworks, but the one aspect I want to highlight, is the governance issue for AI Agents, also known as Visibility in some circles. This is an absolutely critical step in safe, ethical and scalable enterprise AI Agents. Once the industry has settled some of these critical challenges, we will see the market turn to truly scaling AI Agents. I think this future moment is captured well by Workday’s CEO Carl Eschenbach in a February 2025 release of Workday Agent System of Record. When talking about AI Agents, he said:

“The true power of agents is when they become role based, and these role-based agents will maybe start out with a skill, but they will, over time, have many skills, and this is how we will truly unlock the power of AI.”

 Check out my other short articles on AI here:

You can't take the ART out of ARTificial Intelligence

Yes, Leadership Principles Still Apply to Managing a Company with An Agentic Workforce

The Unforeseen Value of AI Agents for HR Leaders