AI agent platforms: architecture and orchestration

Hello architecture enthusiasts! We're excited to announce our next meetup exploring the emerging world of AI agent platforms and orchestration! Join us for an evening where we'll tackle the architectural challenges of moving from standalone AI models to connected, intelligent ecosystems. Makan will share insights on how a capability mesh could solve today's AI orchestration challenges, while Pascal will dive into the practical patterns, protocols, and standards for building AI agent platforms.

Come hungry for both knowledge and food — we've got plenty of both, plus drinks and great networking opportunities at OpenValue's Utrecht office.

Both sessions will be presented in English. RSVP now for an evening of learning and connection!

Getting Here

Directions: https://www.openvalue.eu/route/#utrecht. Parking is free. The office is within walking distance of Utrecht Zuilen train station.

Agenda

17:30 Doors open
18:00 Food
18:30 Talk 1: Capability Mesh: Enabling Orchestration through Semantic Integration
19:30 Break
19:45 Talk 2: Building AI Agent Platforms: Architecture, Communication, and Standards
20:45 Drinks, fun, networking

Talk 1: Capability Mesh: Enabling Orchestration through Semantic Integration
Makan Sepehrifar, Principal Engineer at Code Nomads

Ten years ago, microservices chaos led to "Service Mesh" solving the discoverability problem by replacing network endpoints with meaningful references, enabling scalability and observability. It worked brilliantly for connecting services.

Today we're living through the same cycle with Generative AI and Agents, but the chaos is about integration, not networking. The old problems discoverability, scalability, observability are back because orchestrators now pick agents by capability, not name. Plus we face new challenges: explainability of agent actions and the possibility of unintended actions due to the non-deterministic nature of large language models. We need the same evolutionary leap from "Service Mesh" to "Capability Mesh" shifting from technical service contracts to semantic capability integration for autonomous AI.

Key Topics and Take Aways:

• From OpenAPI to MCP: The Protocol Shift in Agentic AI
• MCP Gateways: The New Control Plane
• Capability Hierarchies: Composing Agents at Scale
• Human-on-the-Loop: Redefining Our Role in Agent Systems
• Policy as Code: Governing Autonomous Agents
• Observability & Safety: Tracing Agent Behavior

About Makan

Makan Sepehrifar is a software architect whose fascination with the intricacies of organizational behavior and cognitive psychology has led him to redefine the way technology interfaces with human understanding. Looking at software architecture as a craftsmanship process, he realized that the more abstract you look at it, the more complexities arises. His career has been a testament to his commitment to bridging the gap between technology and human behavior, resulting in innovative software solutions that not only work efficiently but also self-adaptive teams that embrace the mindset of agile.

Talk 2: Building AI Agent Platforms: Architecture, Communication, and Standards
Pascal Wilbrink, Senior Software Engineer at OpenValue

AI systems evolve from standalone models into connected, agent-based ecosystems, new architectural patterns, protocols, and standards are emerging. Concepts such as AGUI, A2A, A2UI, and MCP-based applications are changing how intelligent systems are designed and integrated.

This presentation explains how these approaches affect system structure, responsibilities, and interoperability. Using architectural diagrams, it shows how AI agent applications communicate and collaborate through standardized interfaces, and provides guidance on building scalable and maintainable AI platforms.

About Pascal

Pascal is a senior software engineer at OpenValue with over a decade of experience and a strong interest in applied AI and intelligent automation. He enjoys exploring how technology can make developers’ lives easier and software more resilient. When he’s not experimenting with new ideas or building open-source tools, Pascal likes to learn by doing and turning complex problems into practical solutions.