Event-driven architecture (EDA) has made many processes and technologies more agile, seamless, and efficient — and now, even generative AI stands to benefit. Beyond providing real-time enterprise data to AI models, EDA also supports communication between AI agents.
According to Shawn McAllister, Chief Product Officer and Chief Technology Officer at Solace, AI has evolved from rudimentary, rule-based systems to sophisticated generative AI capable of complex pattern recognition and reasoning.
“Today, we stand on the precipice of a new AI era — agentic AI. These advanced systems can autonomously perceive, decide, and act within their environments, adapting to new situations with minimal human intervention,” he said in an interview with Frontier Enterprise.
EDA and AI
For many organisations, the AI challenge is composed of 80% integration and 20% AI.
On the integration side, real-time awareness of various aspects of the enterprise is crucial to maximising the potential of generative AI, McAllister noted.
“This is where event-driven integration plays a key role in supporting the flow of real-time information business systems into data stores used by AI — such as vector databases — that can then be leveraged by large language models for retrieval augmented generation,” he said.
He also highlighted how AI agents are increasingly working together in an event-driven manner, coordinated by orchestrators to handle complex tasks. This approach, he said, allows for greater adaptability in workflows and reduces the need for rigid, pre-defined processes.
From McAllister’s perspective, the evolution of EDA has been pivotal to these new AI capabilities. Previously, streaming and publish-subscribe models were niche concerns in application processing, used by only a few. Today, they have become mainstream patterns for sharing real-time data and event notifications.
Moreover, the era of a single, fully integrated tech stack across the enterprise is a thing of the past. Instead, software as a service has introduced bespoke integrations, significantly increasing complexity.
“This integration challenge can be addressed by applying an event-driven approach to create event-driven integration — decreasing coupling between these integration components so data can flow quickly and robustly,” he said.
Broken links
One challenge enterprises often face when integrating EDA is that not all sources of events are event-enabled.
“Some applications, such as SAP S/4 and Salesforce, can natively emit events as changes occur. For systems that do not, alternative techniques — such as polling APIs to detect changes or using change data capture to track modifications — are required to generate reusable events,” McAllister said.
He also remarked that integrating with target applications is often easier, as most have APIs. In such cases, the challenge lies in transforming data into a format that the target system can process.
“Without some type of catalog or event portal, enterprises struggle to identify and integrate the events they need across different systems,” he continued.
Beyond technical challenges, McAllister pointed out that organisations also need to shift their mindset from actively requesting data — such as polling or making API calls — to enabling systems to stream changes as they happen.
“This means allowing the free flow of valuable data and treating changes as events in real time. By adopting this mindset, systems can stay updated without unnecessary bottlenecks. Instead of constantly requesting data, we can simply ‘tune in’ to events and integrate them into target systems,” he explained.
Meanwhile, traditional integration methods — often built on centralised and monolithic architectures — struggle to handle the increasing number of applications deployed across hybrid and multi-cloud environments. According to McAllister, these centralised approaches tightly couple connectors, transformations, and transactional contexts within a single runtime, which forces data to flow in a synchronous manner.
“This often leads to bottlenecks and systems that cannot efficiently handle traffic bursts or surges in data volume. As new integrations are added or removed, data distribution is constantly shifting. When source systems are tightly coupled to target systems, organisations end up with a ‘spaghetti architecture’ — an unstructured, complex web of point-to-point integrations,” he said.
To address this, organisations can adopt event-driven approaches that distribute and filter data to target systems, reducing complexity in routing and delivery. Additionally, event management platforms can help streamline the design and oversight of event streams, making it easier to connect publishers and subscribers.
Further uses
As enterprises continue adopting EDA, McAllister expects event-based API management to emerge as a key trend. Whether in banking, aviation, or logistics, the demand is shifting towards real-time data exchange rather than relying solely on traditional APIs.
“To that end, API management vendors are recognising this shift and exploring ways to enable stakeholders and developers to access these streams of information. Many enterprises already have developer portals and marketplaces for APIs, and similar considerations are now being raised for event streams,” he observed.
Returning to AI, McAllister noted that enterprises still face challenges with real-time data access, data silos, and scalability.
“By combining EDA with integration strategies, organisations can create a unified layer that routes real-time information across applications, devices, and systems, regardless of location. This data can then be fed into AI systems to provide real-time context, improving accuracy and enabling smarter outcomes. Additionally, multi-agent AI systems can work together in an event-driven manner, much like microservices. As AI continues to evolve, event-driven platforms will play an increasingly critical role in the future of AI-powered solutions,” he concluded.
