OpenAI's Codex agents are transforming how employees tackle complex tasks, shifting from fleeting chatbot interactions to sustained, multi-step problem-solving. Employees now spend over 85% of their weekly output tokens using Codex, with 99.8% of tokens in some departments tied to agentic workflows. Non-technical teams like Legal and Recruiting now rely on Codex for more than 85% of their work, reflecting a 137x surge in non-developer adoption since August 2025. By May 2026, 70.2% of users requested tasks equivalent to over an hour of human effort, showcasing agents' growing capacity for deep, iterative work.

Codex’s evolution mirrors expanding technical capabilities. As models strengthened and features expanded, the tool took on tasks like automation, data transformation, and debugging—once reserved for engineers. OpenAI’s Research department saw a 56x usage increase in six months, while Legal grew 13x. Employees now orchestrate parallel agents mid-day, with top users generating 60+ hours of agentic work daily by June 2026. This shift underscores agents’ role in bridging technical and non-technical skill gaps, enabling cross-functional workflows.

The trend signals a broader economic shift. Non-developers, particularly in business operations, now handle engineering tasks in 25% of their Codex output, reducing reliance on specialized expertise. This democratizes technical work, reshaping job roles and skill priorities. OpenAI’s data highlights a frontier adoption pattern: as tools improve, users tackle longer, more complex tasks. The change isn’t just operational—it redefines labor market dynamics, hinting at AI’s potential to expand productivity frontiers across industries.