TECHNICAL DEEP DIVE & DEMONSTRATION: DESIGNING & IMPLEMENTING ENTERPRISE AI AGENT ARCHITECTURES - FROM PROTOTYPE TO PRODUCTION

Early-stage Agentic AI systems are delivering impressive initial results, but many are breaking down when exposed to real enterprise environments. What appears to work in controlled settings quickly becomes unstable when agents are required to interact with fragmented data systems, invoke multiple tools, and operate across longer, multi-step workflows. Outputs become inconsistent, failure points multiply, and debugging becomes increasingly opaque- leaving engineering and architecture teams stuck between promising prototypes and unreliable production systems.

In response, organizations are beginning to formalize how agent-based systems are designed, structured, and governed - moving beyond prompt-level experimentation toward full architectural frameworks that account for orchestration, memory, tool usage, and failure handling. This includes rethinking how agents manage state, how decisions are evaluated, and how systems are instrumented for observability and control, enabling more predictable, scalable, and production-ready implementations.

Please Note - this is not a passive event. Throughout - and at the end of the speakers’ presentation, all attendees are expected to actively participate in the discussion and contribute real-world challenges and solutions.

Topics of discussion will include, yet will not be limited to:

• Designing multi-agent systems to handle complex workflows, including orchestration logic, tool chaining, and state management across tasks
• Identifying where agent-based systems break down in production environments, including failure modes such as hallucinations, tool misuse, latency issues, and inconsistent outputs - and implementing mitigation approaches
• Integrating enterprise data systems into agent workflows, including retrieval strategies, API interaction patterns, and data access constraints
• Implementing guardrails, observability, and evaluation frameworks to improve reliability, traceability, and system performance over time

Santhosh Vijayabaskar, Director, Intelligent Automation & Process Excellence, FIDELITY INVESTMENTS