OpenAI published research on June 25, 2026, detailing the economic transition toward agentic AI tools within their internal operations. The study highlights the shift from short, self-contained chatbot interactions to independent agents capable of long-horizon tasks, such as orchestrating tool calls and iterating toward solutions over minutes or hours. At OpenAI, Codex adoption surged between August 2025 and May 2026, evolving from a secondary tool to the primary resource for every department, including Legal, Finance, and Recruiting. By early June 2026, Codex usage accounted for 99.8% of weekly output tokens within the organization.

User behavior data confirms a trend toward complex, long-duration tasks. By May 2026, 80.6% of individual users performed tasks requiring over 30 minutes of human effort, 70.2% handled tasks exceeding one hour, and 25.6% executed tasks requiring more than eight hours. This expansion of capabilities has enabled non-technical staff to perform cross-functional work, such as coding, data transformation, and structured analysis, effectively bypassing previous technical bottlenecks.

Adoption growth among non-developers significantly outpaced engineers. Between August 2025 and June 2026, non-developer individual users grew 137x, while organizational users grew 189x. Within OpenAI, engineers were the early adopters, reaching majority usage by December 2025; lawyers and recruiters followed, crossing the majority threshold by April 2026. Currently, the average engineer generates 99% of their output tokens via Codex, while non-technical staff generate more than 85% of their tokens using the tool. This shift demonstrates that providing broad, low-friction access to agentic tools allows workers to undertake more complex and cross-functional responsibilities.

OpenAI published research on June 25, 2026, detailing the economic transition toward agentic AI tools within their internal operations. The study highlights the shift from short, self-contained chatbot interactions to independent agents capable of long-horizon tasks, such as orchestrating tool calls and iterating toward solutions over minutes or hours. At OpenAI, Codex adoption surged between August 2025 and May 2026, evolving from a secondary tool to the primary resource for every department, including Legal, Finance, and Recruiting. By early June 2026, Codex usage accounted for 99.8% of weekly output tokens within the organization.

User behavior data confirms a trend toward complex, long-duration tasks. By May 2026, 80.6% of individual users performed tasks requiring over 30 minutes of human effort, 70.2% handled tasks exceeding one hour, and 25.6% executed tasks requiring more than eight hours. This expansion of capabilities has enabled non-technical staff to perform cross-functional work, such as coding, data transformation, and structured analysis, effectively bypassing previous technical bottlenecks.

Adoption growth among non-developers significantly outpaced engineers. Between August 2025 and June 2026, non-developer individual users grew 137x, while organizational users grew 189x. Within OpenAI, engineers were the early adopters, reaching majority usage by December 2025; lawyers and recruiters followed, crossing the majority threshold by April 2026. Currently, the average engineer generates 99% of their output tokens via Codex, while non-technical staff generate more than 85% of their tokens using the tool. This shift demonstrates that providing broad, low-friction access to agentic tools allows workers to undertake more complex and cross-functional responsibilities.