OpenAI's latest research reveals that Codex, initially built as a coding assistant, has evolved into a broad workplace AI system used across finance, legal, recruiting, and other departments to delegate complex, multi-step work tasks. A new study titled "The Shift to Agentic AI: Evidence from Codex" shows that non-developer adoption is accelerating faster than expected, with usage patterns suggesting a fundamental shift in how knowledge workers interact with AI.

What Is Agentic AI, and How Does It Differ from ChatGPT?

The research distinguishes between conversational AI, like ChatGPT, and agentic AI systems like Codex. While ChatGPT primarily answers questions and provides information, agentic systems let users delegate multi-step tasks that require the AI to inspect files, execute commands, and create or modify artifacts. OpenAI frames this shift as moving "from single interactions to delegated, long-horizon tasks".

The paper notes that Codex users increasingly ask the system to perform concrete work, such as debugging code, refactoring applications, drafting documents, analyzing data, and coordinating communication. "These activities are better understood as production than as consultation," the researchers stated, emphasizing that users are asking Codex to do work, not only to provide advice.

How Rapidly Is Codex Adoption Growing Across Different User Groups?

OpenAI studied three distinct populations: individual users on personal plans, organizational users on Business and Enterprise plans, and OpenAI's own workforce. The adoption patterns reveal striking differences in how quickly each group has embraced agentic AI.

• Individual Users: Non-developer adoption rose 137-fold since August 2025, with Codex accounting for 16.5% of output tokens generated by this group as of June 11, 2026.
• Organizational Users: Non-developer adoption increased 189-fold over the same period, with Codex representing 63.3% of output tokens among enterprise and business plan customers.
• OpenAI Employees: The most dramatic shift occurred internally, where Codex accounted for 99.8% of output tokens generated by workers across Codex and ChatGPT combined, effectively replacing ChatGPT as the primary work interface.

The company emphasized that OpenAI's internal adoption is not representative of typical organizations. The research notes that OpenAI employees benefit from familiarity with frontier models, low marginal usage costs, high organizational buy-in, and workflows closely aligned with the systems being developed.

Which Departments Are Adopting Codex Most Aggressively?

One of the study's most surprising findings is the rapid adoption of Codex among non-technical departments. Inside OpenAI, every department, including Legal, Finance, and Recruiting, now uses Codex as its primary AI tool for work. The average lawyer or recruiter at the company now generates more than 85% of their output tokens on Codex.

Legal, Finance, and Recruiting crossed into majority Codex use around April 2026, several months after engineering had moved first. This timeline suggests that once agentic AI workflows become established, adoption spreads rapidly across organizational functions. The research categorizes non-developer tasks into data analysis, research, knowledge artifacts, collaboration, and business-function workflows, demonstrating that agentic AI extends far beyond software development.

How Complex Are the Tasks Users Delegate to Codex?

The research also examined task complexity and duration, finding that heavy Codex users delegate increasingly sophisticated work. By May 2026, 80.6% of sampled individual users had made at least one Codex request estimated to exceed 30 minutes of human work, 70.2% had made one exceeding one hour, and 25.6% had made at least one request exceeding eight hours.

OpenAI notes these duration figures are model-estimated and should be treated as directional rather than exact. However, the broader pattern is clear: intensive users operate Codex differently from occasional users. Heavy users are more likely to run longer and more complex tasks, operate multiple agents concurrently, and use advanced skills. For the most intensive OpenAI users, Codex functions "less an assistant answering requests and more like a workflow system".

Steps to Understanding Agentic AI's Role in Your Organization

• Assess Current Workflows: Identify tasks that require multiple steps, file inspection, or command execution, as these are candidates for agentic AI delegation rather than conversational assistance.
• Evaluate Adoption Readiness: Consider whether your organization has adequate training, knowledge sharing, and buy-in from leadership, factors that OpenAI identified as critical to successful agentic AI deployment.
• Monitor Cross-Departmental Use: Track adoption patterns across departments, not just engineering, since OpenAI's research shows non-developer adoption is accelerating and may reveal new use cases within your organization.
• Plan for Concurrent Task Management: As users become more sophisticated, prepare infrastructure and workflows to support multiple agents running in parallel, a pattern the research identified among intensive users.

What Does This Mean for the Future of Work?

OpenAI's research suggests that the distinction between conversational AI and agentic AI will become increasingly important as organizations adopt frontier models. The shift from asking AI for advice to delegating work represents a fundamental change in how knowledge workers interact with technology. The paper documents adoption, user groups, task categories, task complexity, and output-token patterns, though it does not directly measure productivity gains, software-quality improvements, or business outcomes.

The research was published on June 25, 2026, and reflects data collected through mid-June. As agentic AI systems mature and become more widely available, organizations will need to rethink workflows, training, and infrastructure to support delegated, long-horizon tasks across departments. The rapid adoption among non-developers at OpenAI suggests that the agentic AI shift is not limited to software engineering but represents a broader transformation in how work gets done.