The most important thing about QCon AI New York's December program is not the star power of its speaker list, though seeing engineers from LinkedIn, Netflix, DoorDash, Figma, and Google DeepMind on a single stage is impressive. What matters is the subject matter they are choosing to discuss. These teams are not here to pitch moonshots or theoretical architectures. They are here to explain how they keep AI systems running in production, agent permissions, eval loops, inference cost, code review, and team adoption. That shift from "what AI can do" to "how we make it work reliably" is the story we should all be paying attention to.
This focus on production reality is timely, especially when you consider the regulatory environment taking shape. New York's AI Safety Law Brings New Transparency Requirements for Developers will require major developers to document how their systems behave. The engineers at QCon are already building the guardrails and eval loops that will make that compliance plausible. Meanwhile, OpenAI gives businesses new tools to help ChatGPT guide discovery, signaling that even the largest labs are racing to make their models more controllable in enterprise contexts. The connection is clear: the industry is moving from raw capability to operational discipline. The companies that master production AI, not just training models, will define the next phase of this technology.
For our readers, the practical takeaway is direct. If you are building with AI today, the hardest problems are not about choosing a model. They are about how you constrain it, how you evaluate its outputs at scale, and how you get your team to adopt it without breaking your workflows. The QCon program confirms that these are the universal challenges. An engineer at Netflix faces the same guardrail questions as one at Honeycomb. A team at DoorDash debugging inference cost has the same math problem as a team at Pinterest. The specificity of the talks, agent permissions, eval loops, code review, gives you a checklist of your own blind spots. If you are not thinking about how your AI agents are permissioned, or how you will measure their output drift over time, you are behind the practices these teams are already running in production.
One detail to watch is the talk on team adoption. It is easy to fixate on the technology, but the human layer, getting engineers to trust AI code review, getting product managers to accept agent-driven workflows, is often the bottleneck. OpenAI's latest move signals a deeper interest in travel data reminds us that these tools are being embedded into specific verticals, which means adoption is not just a technical problem but a cultural one. The companies that succeed will be those that treat their teams as part of the production system, not just users of it.