Performance at AI Scale: What We Learned at monday.com

For years, performance engineering at monday.com focused on both client and server: identifying shifting bottlenecks, optimizing data flow, and ensuring responsiveness at scale. Our core product — Boards — underwent more than seven years of deep performance work, including architectural changes, caching strategies, data modeling, and query optimization. Along the way, we learned a critical lesson: bottlenecks are never fixed. As usage patterns evolve, performance pressure moves between the server, the network, and the client, and understanding these shifts is essential to delivering a smooth experience.
When AI became a first-class citizen in the product, we faced a new challenge: applying everything we had learned to a far more dynamic and unpredictable environment. AI-driven clients generate highly variable, data-heavy usage patterns that no traditional benchmarks could predict. Performance was no longer just about the server or internal clients — it became a problem of translating years of internal optimizations into a system that could handle external, rapidly changing clients while maintaining speed and reliability. In just three months, we had to enable custom client-side applications for users, building performance guardrails that could adapt to unknown usage patterns while preventing overload and maintaining responsiveness.
This talk explores the lessons learned from tackling performance in this AI-driven world. We’ll discuss how traditional assumptions break down when client behavior is no longer controlled, why guardrails are as important as raw speed, and how to quickly apply deep platform knowledge to new and unpredictable scenarios. We’ll share what worked, what didn’t, and the design decisions we had to rethink to succeed.
This session is for engineers and platform leaders building APIs or extensible products at scale. In the AI era, performance is no longer just an optimization problem — it’s a platform design problem.