Scaling ChatGPT Work to 10 Million Users: Insights from OpenAI's Akshay Nathan
In an interview on Latent Space, OpenAI's core product engineering lead Akshay Nathan detailed the technical and product strategies used to scale ChatGPT Work to 10 million users. The discussion covers key architectural components such as memory, subagents, OpenClaw integration, and no-code features. Scaling AI products to millions of users requires robust engineering architectures that handle complex agentic workflows and memory management. Understanding OpenAI's approach provides a blueprint for developers and enterprises building large-scale, production-ready AI applications. The architecture leverages subagents to separate reasoning from execution, ensuring fresh context and specialized task handling. Additionally, it integrates OpenClaw, an agentic interface for autonomous workflows, and supports no-code integrations to make artificial general intelligence (AGI) more accessible.
## BACKGROUND
Subagent architectures orchestrate multiple specialized AI agents under a main coordinator to handle tasks that exceed a single model's context window or require parallel execution. OpenClaw is an open-source AI assistant framework designed to run locally and connect with external large language models like OpenAI's GPT series to automate workflows.