A new research paper from OpenAI suggests AI agents are moving beyond coding in large enterprises, taking more complex workplace tasks in business functions including legal, finance and recruiting.
According to The Shift to Agentic AI: Evidence from Codex, the company’s Codex programming model has increasingly displaced chatbot-style interactions for many users as the platform has gained the ability to execute longer-running tasks across multiple tools and systems. The report also found growing use among non-technical workers. Since August 2025, non-developer organizational users increased by 189 times, while non-developer individual users grew 137-fold, according to OpenAI’s analysis.
The paper argues that AI agents are shifting from answering isolated questions to completing delegated work that can take hours of human effort. By May 2026, the report said more than 70 percent of sampled individual users had submitted at least one task estimated to require more than one hour of human work.
“As the tools improve, people use them for longer, more complex, and more cross-functional work,” the report’s authors wrote.
OpenAI researchers said the findings illustrate how agentic AI could change enterprise workflows by allowing employees to complete technical and cross-functional work that previously required specialized expertise. The company acknowledged that the research reflects adoption patterns among frontier users rather than the broader labor market.