QCon AI NY, Production Engineers Speak on Real-World AI

The technical conference 'QCon AI New York 2026' has unveiled its full program for its December 15–16, 2026 event. Engineers from LinkedIn, Netflix, DoorDash, Figma, Pinterest, Honeycomb, Harvey, Google DeepMind, and OpenAI will present on AI systems currently operating within their respective organizations.

The discussion surrounding AI is rapidly shifting from 'What can we build?' to 'How do we operate it?' Unlike previous conferences dominated by research-stage presentations, this event puts engineers with hands-on production experience front and center. Regardless of company size or industry, the conference serves as a venue for sharing challenges encountered when integrating AI into actual services.

The announced presentation themes cover five areas: AI agent permission settings and guardrails (constraints to maintain safe operation), evaluation loops (evals), inference costs, AI-assisted code review, and team adoption processes. All of these are part of the challenge set required not to 'build' AI but to 'keep running' it—addressing problems that field engineers confront daily.

AI agent guardrails and evaluation loops are particularly noteworthy themes gaining attention lately. Agents refer to AI systems that autonomously complete tasks based on instructions. Running such systems safely in production environments demands critical decisions: how much authority to grant and how to design mechanisms for continuous measurement of whether operations align with intent. How multiple major tech companies approach these questions holds significant value as industry-wide knowledge.

Inference cost is another crucial topic. The computational resource expenses incurred each time an AI model is called become directly tied to business outcomes as user scale grows. How large-scale services like Netflix and DoorDash manage these costs could serve as valuable reference cases for enterprises facing similar challenges.

The sessions on AI-assisted code review and team adoption shine light on issues separate from pure technical dimensions. Even excellent AI tools are meaningless if development teams cannot effectively use them. The question of how engineer organizations embrace change and weave it into practice aligns with organizational challenges many companies currently face.

The public sharing of operational knowledge from production AI systems holds significant meaning for the entire industry. As insights gained through trial and error by individual companies accumulate horizontally, there is potential for the overall standard of safe and sustainable AI application to be raised from the ground up. Going forward, attention is likely to focus on the details of presentations and the concrete design and operational insights offered by participating companies.