Artificial intelligence is moving beyond chatbots and search assistants toward agentic AI—systems that can independently carry out complex tasks on a user's behalf. OpenAI’s Codex platform is emerging as one of the clearest examples of this transition, allowing users to delegate substantial work rather than simply ask questions. Researchers from OpenAI, Columbia University, Duke University, and the University of Pennsylvania found that usage of Codex is growing rapidly, suggesting that AI agents are beginning to gain traction in real-world workflows.
One of the most striking findings is that many Codex tasks represent significant amounts of work. In a sample of users who allowed their interactions to be studied, more than 80% submitted at least one request estimated to be equivalent to over 30 minutes of work for an experienced human. While the total number of Codex users remains relatively small compared with ChatGPT, those who adopt the platform tend to use it extensively, indicating that AI agents may be delivering meaningful value to early adopters.
The report also reveals that non-developers are the fastest-growing user segment, even though software development remains Codex's primary application. Earlier OpenAI data showed that knowledge workers—including analysts, marketers, and operations professionals—are adopting Codex at a faster rate than programmers. This suggests AI agents are evolving from specialized coding tools into broader workplace assistants capable of handling documents, research, scheduling, analysis, and other professional tasks.
The broader implication is that 2026 may be remembered as the year AI shifted from conversation to delegation. Experts cited by Axios argue that agents reduce the "psychological cost of action" by helping users move from ideas to execution more easily. While most AI users still rely primarily on traditional chatbots, the rapid growth of platforms like Codex suggests that autonomous task completion—not just information generation—could become the next major phase of AI adoption.
