Amid potential challenges in implementing agentic AI adoption, one event-driven approach could facilitate real-time data flows and autonomous decision-making across industries
Agentic AI — an AI system that does not just follow pre-programmed instructions but thinks on its own feet, is capable of making autonomous decisions and adapting to new situations.
However, successful adoption of agentic AI hinges on organizations’ willingness to embrace change and integrate these technologies into their operations.
To do that, they first need to understand the value of AI for their business.
A primer on agentic AI benefits
Agentic AI goes far beyond simple question-answering on a subject for which a large language model has been trained. Instead, it operates as a sophisticated software pattern that orchestrates multiple LLMs and services, also known as “agents”, to perform more complex tasks and reasoning autonomously.
Initially, such systems were limited to rule-based tasks. They have now steadily advanced into sophisticated, multimodal agents. These agents possess the ability to process and integrate information from diverse sources, including text, images, and audio. This multimodality empowers AI agents with reasoning capabilities that can interact in ways that can almost simulate human understanding.
From streamlining customer communications and optimizing real-time inventory decisions to enhancing fleet management, agentic AI could revolutionize day-to-day operations across multiple industries.
Downsides to agentic AI expectations
With that said, getting out of the pilot phase and into everyday work applications is proving to be the biggest hurdle for any AI project.
With anecdotal evidence showing that AI projects have a failure rate as high as 80%, and many APAC IT leaders still struggling to integrate data across their systems, implementing agentic AI for accurate and effective outcomes can be challenging.
In addition, other studies have suggested that many AI projects fail to scale due to legacy architecture dependencies, and cost and performance challenges in scaling something so complex and unstructured.
Even when projects do get up and running, data quality, governance, security, and tech workflow integration hurdles may remain.
Understanding the “event mesh” concept
At the heart of AI implementation challenges lies a critical deficiency: the absence of real-time, contextual information flow.
Traditional batch processing and static data models still in use by many organizations fall short of providing dynamic business environments where critical decisions need to be made in split-seconds.
An event-driven architecture (EDA) paradigm, the event mesh can power enterprise AI into a real-time, context-aware solution. It provides the decoupling needed for rapid development and change, and it delivers on the EDA that allows for managing rate mismatch, supporting different applications with messaging patterns, and delivering the efficiency needed to scale horizontally and vertically.
Integrating EDA with agentic AI
When the architectural pattern enabled by the event mesh is applied across agentic AI use cases, a flexible, real-time data distribution network is created, that enables various AI models to access and react to relevant data streams instantly.
While an event mesh enables real-time data flow and dynamic routing across the enterprise, an agent mesh takes this further by introducing intelligent agents that can autonomously reason about, and act on, this information flow.
Such a framework may offer potential benefits such as:
- The building of a network of AI agents overseen and controlled by a dynamic orchestration layer, allowing complex tasks to use multiple agents and have their results knitted together in a data management system.
- In this context, mesh gateways can be set up to serve as interface points for the agent network, allowing various use cases to access the system. Each mesh gateway can provide a unique interface with its own input methods and authorization rules. These gateways enable the AI agent network to interact with different data sources, applications, and user interfaces, facilitating the integration of AI capabilities across diverse systems and use cases.
Essentially, when powered by an event mesh, potentially-more-advanced autonomous AI systems can be built that can manage requests to deliver the desired results based on unstructured inputs.
A composable, flexible AI framework
As a plug-and-play approach, implementing EDA to power agentic AI is not intrusive to an organization’s existing application stack and agentic AI framework.
Organizations can start small with one or two use cases and then, over time, add more agents and gateways to increase the system’s capabilities. Then, with orchestration and built-in access control of all agents and actions in the system, one framework can be used and re-used for many use cases — each providing different interfaces and access control that is governed by enterprise-grade security.
Furthermore, the decoupled nature of an event-driven agentic AI framework can allow organizations to easily update, replace, or add new AI models and data sources without disrupting existing systems. This is especially crucial for staying current with AI advancements.
Editor’s note: For a deeper exploration of the advantages, challenges and alternatives of EDA in AI systems, readers can refer to this study, or this and other reports as appropriate.