The Future of Agentic AI Ecosystems
Image: Depositphotos
Source: Irving Wladawsky-Berger
Wikipedia defines Agentic AI as “a class of artificial intelligence that focuses on autonomous systems that can make decisions and perform tasks without human intervention.” A recent IBM website, “What is Agentic AI,” adds that: “Agentic systems provide the flexibility of LLMs, which can generate responses or actions based on nuanced, context-dependent understanding, with the structured, deterministic and reliable features of traditional programming.”
“Broadly speaking, agentic systems refer to digital systems that can independently interact in a dynamic world,” explained a McKinsey report published in July of 2024. “While versions of these software systems have existed for years, the natural-language capabilities of gen AI unveil new possibilities, enabling systems that can plan their actions, use online tools to complete those tasks, collaborate with other agents and people, and learn to improve their performance. Gen AI agents eventually could act as skilled virtual coworkers, working with humans in a seamless and natural manner. … In short, the technology is moving from thought to action.”
“Billions of dollars of investment by some of the largest firms on the planet are flowing into tools that will make it easy to build autonomous agents,” wrote senior technology executive Eric Broda in “Agentic Mesh: The Future of Generative AI-Enabled Autonomous Agent Ecosystems,” a November 2024 article in the Medium platform. “And if this huge investment, and the recent headlines, are any indication, we will soon have many, many autonomous agents collaborating in a dynamic ecosystem.”
The key questions will not be “how to build autonomous agents,” he added, “but rather, how do we manage this burgeoning ecosystems of autonomous agents. How does one find an autonomous agent that does what we want? How does one interact with an autonomous agent? And if we want to transact with an autonomous agent, how does that happen? And how does it happen safely?”
In his Medium article, Broda discussed the evolution of agentic systems, from simple agents to what he calls Agentic Mesh ecosystems. Simple agents use LLMs to interpret and respond to user instructions and to perform specific actions by following straightforward commands within well-defined boundaries with minimal human intervention. In multi-agent systems, multiple agents, each with distinct roles and capabilities, begin working together to achieve more complex goals collaboratively with limited human interventions. Over time, communities of agents will start to emerge, that is, networks of interoperable agents that work together on complex projects and create solutions in real time.
Such agentic ecosystems can unlock significant value for businesses based on their potential to automate the highly variable number of relatively simple variations in complex applications that have historically been difficult to address in a cost- or time-efficient manner. For example, a virtual assistant could organize a complex business trip encompassing different airlines, flights, restaurant reservations and business meeting each of which must be handled across different online platforms. While each individual step is relatively simple, the coordination of the multiple steps involved in the overall trip can be quite complex and hard to automate.
Programmers are evolving from their traditional application development roles. They’re now becoming high level application engineers whose job is to design and clearly specify the overall application they're developing including the various human assistants and AI agents needed to get the work done. Their job is to define the goals and actions the agents should follow under a variety of conditions using LLMs and other AI tools, as well as to make sure that the overall application is working as intended.
“As GenAI capabilities grow exponentially — and their cost shrink exponentially — we foresee a vast and diverse ecosystem of GenAI-powered agents that are not only fit-for-purpose but with a wide range of costs commensurate with their scope and value,” wrote Broda. “We expect to see agents using LLMs that are big, small, and everything in between.” Agents will become specialists — some with industry specific expertise in their fields, while other agents may focus on orchestration, execution planning, governance and compliance.
“However, the underlying expectation for all agents is that they are trustworthy, they are safe, they are reliable, and they act as expected,” he added. “Today, a user of ChatGPT is literally the human in the loop. However, when agents handle operational tasks independently, we need to make it easy to understand not just agent capabilities but also their operational policies, track record in achieving outcomes, any published feedback from people using them, as well as the availability of third-party audit and certification results.” In addition, as AI technologies continue to improve rapidly, we must be very careful with how much control, autonomy and power we hand over to AI systems before we understand how well they work and how they can go wrong.
Over time, Broda envisions the creation of what he calls an Agentic Mesh, which he defines “as an interconnected ecosystem where federated autonomous agents and people initiate and complete work together.” The Agentic Mesh is designed to support agent collaboration, foster trust, maintain a significant degree of autonomy and make it easy for agents to find each other, and safely collaborate, interact, and transact.
To operate effectively and efficiently in an Agentic Mesh, agents must be designed with six key characteristics that define their behavior and scope:
Autonomous. An agent can act independently within the limits set by its owner and purpose. “Autonomy allows agents to make decisions and take actions without constant supervision, which is essential for scalability and real-time operations. However, by defining clear boundaries, agents can adapt and act freely while remaining aligned with their goals and ownership rules.”
Purposeful. An agent’s purpose defines the boundaries within which it must operate, ensuring that its actions remain relevant and aligned with specific goals. The agent's purpose should be published, transparent, and easily available. “A well-defined purpose keeps the agent focused and provides the high-level boundaries that enable policy enforcement.”
Accountable. Each agent requires an accountable owner that not only creates the Agent but is also responsible for its performance and outcomes. “Ownership is the basis upon which governance is built and allows for the delegation and enforcement of authority and control. And, obviously, information about the owner is a crucial proxy in determining the trustworthiness of an agent.”
Intelligent. Agents use LLMs to permit interaction via natural language and to determine execution paths. They also use fit-for-purpose LLMs — big or small, general or specialized, costly or inexpensive — to complete their task in an effective manner.
Discoverable. Agents must be discoverable by both users and other agents. “Agent discovery is the process of locating an agent within an Agentic Mesh based on specific criteria, enabled by the agent’s registration information (purpose, ownership, policies, etc).”
Trustworthy. Transparent policies codify an agent’s purpose allowing it to be verified, certified, and audited. “An agent must exhibit trustworthiness by behaving consistently, predictably, and within the bounds of its purpose. Trustworthy agents inspire confidence in users, owners, and other systems by reliably fulfilling their roles and adhering to expected standards. Trustworthiness also includes providing and publishing (in the Registry/Marketplace) proof of compliance with ethical guidelines and legal requirements. Providing audit trails and error-handling mechanisms further reinforces trust, ensuring that agents maintain reliability even under unforeseen conditions.”
“Agentic Mesh ecosystems solve a practical problem in a large agent ecosystem: it lets agents find each other, collaborate, interact, and transact,” wrote Broda in conclusion. “As agents become more capable, reliable, and trustworthy, it is likely inevitable that businesses will rely on agents to automate tasks, respond dynamically to changes, and unlock new efficiencies.”
Irving Wladawsky-Berger
Irving Wladawsky-Berger, PhD., is a Research Affiliate at MIT's Sloan School of Management and at Cybersecurity at MIT Sloan (CAMS) and Fellow of the Initiative on the Digital Economy, of MIT Connection Science, and of the Stanford Digital Economy Lab.
Visit Irving on LinkedIn.