OpenAI's Internal AI Agent Serves 4,000 Staff
An internal AI data agent built by just two engineers at OpenAI now supports over 4,000 employees with complex data tasks. The company says the tool, which automates multi-step workflows, can be replicated by other organizations, showcasing the rapid maturation of bespoke AI assistants for enterprise use.
The internal data agent manages a staggering 600 petabytes of data spread across 70,000 datasets, serving over 3,500 employees. This allows teams in engineering, finance, and research to get answers from massive datasets in minutes instead of days by using natural language. Built on OpenAI's own GPT-5.2 and Codex models, the agent doesn't just write SQL queries; it writes the *right* ones. It can handle ambiguous questions, like "customer satisfaction in Europe," by knowing which internal metrics and tables are the correct ones to use for that specific query. A key to its success is a six-layer context system that grounds it in organizational reality. The agent understands not just data schemas but also how tables are typically used, incorporates human annotations, and extracts business logic directly from internal software pipeline code. The system also taps directly into OpenAI's institutional knowledge by accessing and embedding information from Slack, Google Docs, and Notion. This allows it to understand the context behind the data, such as product launches or reliability incidents, connecting the "what" with the "why." If a query fails or returns strange results, the agent can autonomously diagnose the error, adjust its approach, and try again without the user even noticing the initial failure. When a human does make a correction, the agent's memory system learns from it, ensuring the same mistake isn't repeated for any other user in the company. The tool is accessible to employees directly within their existing workflows, including Slack, web interfaces, and developer IDEs. Security is handled by inheriting the existing data permissions of the user, meaning employees can only query data they already have access to.
