Speaker of TestIstanbul 2025
Senior Staff Performance Engineer @ServiceNow
Speech: Performance Testing AI Agents: Strategies for Scalable and Reliable Enterprise Systems
As AI-driven agents become increasingly embedded in enterprise applications — from virtual assistants and chatbots to intelligent routing and decision-making engines — ensuring their performance under real-world conditions is critical. Unlike traditional software components, AI agents exhibit non-deterministic behavior, dynamic learning patterns, and rely heavily on upstream data quality, latency-sensitive APIs, and external model inference services. These characteristics introduce new challenges in performance engineering that extend beyond throughput and response time.
This talk explores a systematic approach to performance testing for AI agents in production-grade systems. Drawing on real-world case studies, including multi-threaded AI routing in customer service platforms and NLP-driven chatbots, it explores how to build test harnesses that simulate user intent variability, model drift, and token-level latency. Key topics include synthetic data generation for AI load tests, metrics beyond classical SLAs (e.g., intent resolution time, confusion score, inference latency), and strategies for aligning AI observability with performance testing tools like JMeter and custom profilers.
Through practical techniques and architectural insights, this talk aims to equip performance engineers and SREs with a robust framework for validating AI agent scalability, resilience, and responsiveness in modern digital environments.
About Sudhakar Reddy Narra
Sudhakar Reddy Narra is a performance engineering architect with over 17 years of experience in building and scaling high-performance systems across enterprise platforms. His work spans end-to-end performance engineering—including architecture analysis, benchmarking, resource optimization, and system tuning for reliability at scale.
He has played a key role in shaping performance strategies for modern cloud-native and AI-powered applications, focusing on both infrastructure efficiency and user experience. Sudhakar actively contributes to the evolution of performance engineering practices, blending deep technical insight with practical testing methodologies to ensure software systems perform predictably under real-world demands.
