# QCon AI New York 2026: Scaling Production AI Systems Industry leaders share operational strategies for managing AI agents, evaluation pipelines, and inference. - **Source:** [InfoQ](https://brifing.si/source/68ac18af-a4f3-4f7c-9c2c-44d4af8f89dd) - **Published:** 2026-11-11 - **Topic:** [technology](https://brifing.si/c/technology) - **Original article:** - **On Brifing:** https://brifing.si/s/48b6c435-a97a-400f-a90d-75f9b96e1f8c ## Conference Focus on Production QCon AI New York 2026, scheduled for December 15-16, gathers engineers from major tech firms to discuss production AI systems. The agenda covers critical operational challenges, including agent authorization, evaluation pipelines, and large-scale model serving. Practitioners will examine how to transition from initial deployments to robust, scalable AI architectures. ## Identity Challenges for Agents Nancy Wang of 1Password highlights that current identity systems are built for human users with stable roles and bounded sessions. Agents lack these traditional characteristics, necessitating new authorization frameworks. The conference explores how to secure these non-human entities effectively while maintaining accountability within complex enterprise software environments and production workflows. ## Figma and LinkedIn Infrastructure Figma improved agent provisioning latency by 90% using warm pools, raising availability to 99.95%. Meanwhile, LinkedIn manages a Kubernetes ops agent across 500,000 nodes and 5 million pods. Their system incorporates server-side guardrails like rate limits and peer approval to ensure stability during automated production changes and large-scale infrastructure operations. ## Evaluation and Quality Control DoorDash utilizes an LLM judge calibrated against human reviewers to validate changes in their agentic ordering system. By using identical rubrics for both simulated and real sessions, they ensure offline evaluation gains translate into production performance. This approach addresses the common industry struggle of maintaining consistent evaluation metrics over time. ## Netflix Serving Platform Scale Netflix engineers Fan Yang and Rajat Shah detail their unified platform that serves machine learning and generative AI models. After five years of consolidation, the system now handles approximately 1 million inference requests per second across 300 models. The presentation focuses on the operational costs and technical requirements of multi-tenancy. ## AI-Assisted Software Development LinkedIn’s AI code review system processes tens of thousands of pull requests weekly across 10,000 repositories. With a 60% acceptance rate for sampled suggestions, the system employs a validation pipeline to filter findings before developer exposure. This highlights the growing role of AI in streamlining engineering practices and maintaining code quality. ## Honeycomb and Organizational Adoption Honeycomb’s 2x initiative demonstrates the impact of AI on engineering velocity. Peak weekday pull request merges to their main monorepo increased from 30 to 74 between early 2025 and April 2026. This sustained growth reflects successful organizational adoption of AI-native tools and improved internal development workflows for their engineering teams. ## Industry Trends and Participation The conference features insights from Anthropic, Microsoft, and BlackRock on topics like AgentOps and sandbox density. With individual tickets priced at $1,790 until December 8, the event serves as a hub for senior developers. These sessions provide a roadmap for teams navigating the complexities of enterprise-scale AI integration and deployment. ## Sources covering this story - https://www.infoq.com/news/2026/11/qcon-ai-newyork-2026-program/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._