What is Agentic AI and Why Does It Matter

Agentic AI represents the next evolutionary stage of artificial intelligence. While classic large language models like ChatGPT or Claude answer questions based on their training, agentic systems can independently plan, make decisions, and perform actions — without the need for constant human oversight of every step.

According to Gartner, by 2028 at least 15% of daily work decisions will be made autonomously through agentic AI. This is not just about simple tasks. Modern agentic frameworks already combine text, image, and audio inputs and can interact in a way that approaches human understanding of context.

Practical example: an agentic AI system deployed as customer support can, upon receiving a new request, analyze the history of previous tickets, extract relevant product usage information, and autonomously generate an informed response — all within seconds, not hours.

Why 80% of Projects Fail: Integration, Not Intelligence

The biggest paradox of the current AI wave is that the "intelligence" of the models themselves is not the bottleneck. A Harvard Business Review study estimates that up to 80% of AI projects end in failure. The reason? A lack of real-time connection to real business data.

MuleSoft states in its report that 95% of IT leaders in the Asia-Pacific region still struggle with data integration across their systems. Without access to up-to-date data from manufacturing, logistics, CRM, or ERP systems, even the smartest AI model remains blind and unusable.

Edward Funnekotter, Chief Architect and AI Officer at Solace, summarizes it clearly: For many organizations, AI is an 80% integration challenge and 20% AI itself. Traditional batch processing and static data models, which still prevail in companies, are simply not sufficient for a dynamic business environment where decisions must be made in real time.

Event Mesh and Agent Mesh: How It All Works

The key concept promoted by Solace is event mesh — an interconnected network of so-called event brokers that dynamically routes information between applications, devices, and systems across various environments. It functions similarly to a company's central nervous system: when something happens anywhere (an order, a machine breakdown, a change in inventory status), the information immediately reaches where it needs to go.

Building on this foundation is the agent mesh — a framework that allows for the creation of a network of AI agents controlled by a dynamic orchestration layer. Individual agents can autonomously process different parts of complex tasks, and their results are then combined in a central data management system.

"Agent mesh gateways enable access to this system for many different use cases, each with its own type of input interface and permissions," explains Funnekotter. The result is a system that can process unstructured inputs — such as a regular customer chat — and convert them into specific actions across the company's infrastructure.

Why This Matters for Czech Companies Too

Agentic AI with event-driven architecture is not just a topic for tech giants. In the context of the European Union — and with the impending full effectiveness of the AI Act — companies will need not only powerful AI systems but, above all, transparent, auditable, and secure ones.

It is precisely the event-driven and agent mesh architecture that offers advantages resonating with European requirements: built-in access control, traceability of individual agent actions, and the ability to deploy the system gradually — from one or two use cases to complex orchestration across the entire company. For Czech businesses, which often operate with more limited budgets than their Western competitors, this gradual scalability is crucial.

Furthermore, Solace Agent Mesh is available as an open-source framework, which lowers the barrier to experimentation. Czech companies that already work with technologies like Apache Kafka for data streaming can also build agentic AI solutions on similar infrastructure.

From Pilot Project to Production: The Key is in Architecture, Not the Model

One of the main advantages of the event-driven approach is that it is not invasive to the existing application stack. Companies do not have to rewrite their systems from scratch — agent mesh works on a "plug-and-play" principle, allowing for a small deployment to start and gradually adding more agents and input gateways.

Thanks to the orchestration layer with built-in access control, one framework can also be used for many different cases — from a new order to a customer ticket to a chatbot query. Each channel has its own interface and level of authorization, all enterprise-secured.

It is also interesting that a decoupled architecture allows individual AI models and data sources to be easily updated or exchanged without disrupting running systems. At a time when new model versions appear every few weeks, this is a significant competitive advantage.

What the Numbers Say: Investments and Predictions

Singapore has allocated 150 million Singapore dollars (approximately 2.6 billion CZK) this year to the Enterprise Compute initiative, which aims to provide companies with computing capacity specifically for the development of autonomous AI systems. Europe is currently lagging in direct investments in agentic AI infrastructure — while the European AI Act creates a regulatory framework, similarly ambitious subsidy programs for the business sector are missing.

For comparison: the Czech National Recovery Plan envisages investments in digitalization and AI in the order of billions of CZK, but primarily directed towards public administration and research, not direct support for corporate deployment.