Generative artificial intelligence (GenAI) has received no small amount of buzz over the past few years, but the area of greatest interest to many organisations is multi-agent AI. However, making this technology work as promised requires a rethink of system architecture.
Multi-agent AI systems build on agentic AI, which can learn, adapt and act independently. While GenAI is already delivering outstanding results for businesses, agentic AI offers even greater potential by automating complex workflows, making real-time decisions, and coordinating tasks across systems without requiring human intervention.
In a corporate setting, agentic AI may be tasked with things like autonomously reviewing customer claims within a financial services firm, verifying documentation and resolving discrepancies — reducing approval times while increasing customer satisfaction.
Multi-agent AI technology takes these capabilities a step further, giving businesses the power to employ networks of autonomous AI agents, each of which specialise in a particular task, and which work together in a coordinated fashion as part of a larger, intelligent system.
Because each part of a multi-agent AI network operates autonomously, communicates progress, shares a common knowledge base, and coordinates actions, such systems are particularly well suited to handling complex, interdependent tasks that a monolithic single AI agent couldn’t feasibly be built to manage alone.
Communication breakdown
As multi-agent AI systems make their way from proof of concepts to production deployments, organisations are coming up against certain challenges, most notably those that limit scalability. Scaling any kind of distributed system often comes with challenges related to communication.
In the case of multi-agent AI networks, the individual AI agents typically need to exchange information to share context and coordinate tasks if they are to work together properly. However, if communication between each agent is accomplished via direct request and response calls, problems can arise as a result of potential bottlenecks and network rigidity.
The same issues have previously been seen in the area of microservices. Like the agents in a multi-agent AI network, individual microservices often need to communicate with each other to function properly. Microservices relying on direct request and response communication systems such as application programming interfaces (APIs) can result in rigid dependencies.
In an expansive network, if a direct API call from one agent or service to another fails or is delayed, it has the potential to affect the timing or delivery of follow-on communications throughout the entire system. This effect can be a substantial limiting factor when it comes to the scalability and real-time responsiveness of multi-agent AI systems.
The main event
This is where event-based architecture can make a difference. Event-driven architecture is a software design paradigm in which events — state changes or notable occurrences — trigger downstream actions asynchronously. Instead of relying on direct calls between components, event-driven architecture uses event producers (like AI agents), event consumers, and event channels to decouple systems and keep information flowing in real time.
In the realm of microservices, with an event-based approach, each microservice can publish and subscribe to events, reducing dependencies, improving scalability, and increasing resilience to hardware failure or network interruption. Likewise, event-based architecture lets AI agents publish and subscribe to events, giving them the ability to operate independently while staying in sync, no matter their number or how they are arranged to complete tasks.
Within the event-driven model, individual AI agents are designed to emit and listen for events autonomously. In this context, events act as signals that something has happened, allowing agents to respond without needing tightly coupled, point-to-point integrations. Ultimately, this approach ensures agility, scalability, and a more dynamic system.
That’s why event-based architecture offers a solution to the challenge of scale in multi-agent AI systems. Just as the introduction of event-driven architecture has eased the rigidity of microservices, it also enables multi-agent AI systems to operate more dynamically, minimizing points of failure in communication between agents.
Sailing the stream
As event-driven architecture is needed to bring scale to multi-agent AI networks, data streaming technology is a key enabler. For instance, businesses using the popular orchestrator-worker multi-agent pattern, which involves a central orchestrator assigning tasks to worker agents, can use data streaming to make the agents event-driven.
If used in conjunction with an appropriate data streaming platform, multi-agent AI systems can operate in real-time, making informed decisions based on high-quality, discoverable, and continuously updated information. In fact, a data streaming platform can act as both a coordination path for agents as well as a common operating model for data orchestration.
More broadly, data streaming is an essential element for agentic AI systems powering applications such as fraud detection, real-time monitoring or control systems where responsiveness and adaptability are critical. Data streaming has the potential to help companies go big with multi-agent AI networks.
The potential of multi-agent AI systems is too great for such technology to be held back by inappropriate architecture. For businesses building toward a future that will benefit from the fast-evolving capabilities of agentic AI technology, now is the time to rethink how AI agents communicate by bringing an event-based approach to their systems.
- About Andrew Sellers & Sean Falconer
- About Confluent
Andrew Sellers –
Dr. Andrew Sellers leads Confluent’s Technology Strategy Group, a team of research technologists, Field CTOs, and customer advisory leaders supporting product and GTM strategy development, competitive analysis, and thought leadership. Andrew has previously brought several AI-enabled commercial offerings to market as a technology leader, applying data streaming extensively to ensure application data is well contextualized and current. Data streaming commercial products he designed, built, and delivered have generated tens of millions in revenue. He is an author of dozens of peer-reviewed publications and a co-inventor listed on over 100 patents, primarily related to the application of AI tools to problems in cybersecurity and risk management. Andrew received his PhD in computer science from the University of Oxford, where his dissertation investigated the use of AI to automate web-scale data extraction. He was the valedictorian of his class at the U.S. Air Force Academy and a Truman Scholar.
Sean Falconer –
Sean is an AI Entrepreneur in Residence at Confluent where he works on AI strategy and thought leadership. Sean’s been an academic, startup founder, and Googler. He has published works covering a wide range of topics from AI to quantum computing. Sean also hosts the popular engineering podcasts Software Engineering Daily and Software Huddle.
Confluent is pioneering a fundamentally new category of data infrastructure focused on data in motion. Our cloud-native offering is the foundational platform for data in motion – designed to be the intelligent connective tissue enabling real-time data, from multiple sources, to constantly stream across the organization. With Confluent, our customers can meet the new business imperative of delivering rich, digital customer experiences and real-time business operations. Our mission is to help every organization harness data in motion so they can compete and thrive in the modern world.
Techedge AI is a niche publication dedicated to keeping its audience at the forefront of the rapidly evolving AI technology landscape. With a sharp focus on emerging trends, groundbreaking innovations, and expert insights, we cover everything from C-suite interviews and industry news to in-depth articles, podcasts, press releases, and guest posts. Join us as we explore the AI technologies shaping tomorrow’s world.
