The Transition to Agentic Workflows

Agentic AI represents a fundamental shift in how knowledge work is performed. Unlike chatbots, which are limited to short, self-contained interactions, agents can operate independently for extended periods—ranging from minutes to hours—by orchestrating tool calls and iterating toward solutions. Data from OpenAI’s internal usage shows that as these tools become more capable, users move away from simple Q&A toward delegating complex, multi-step tasks.

Internal metrics reveal a significant evolution in how employees interact with AI:

  • Extended Horizons: By May 2026, over 80% of individual users performed tasks estimated to take more than 30 minutes, with 25.6% executing tasks exceeding eight hours of human work.
  • Dominance in Output: Within OpenAI, Codex now accounts for 99.8% of weekly output tokens, with the average worker generating over 85% of their output via agentic tools rather than traditional chatbots.
  • Parallelization: Heavy users at the 99th percentile now generate more than 60 hours of agent work per day, often orchestrating multiple parallel agents to manage complex workflows.

Democratization of Technical Tasks

While agentic tools originated as coding aids for engineers, adoption among non-technical departments (Legal, Finance, Recruiting) has grown exponentially—outpacing developer adoption by a factor of 189x for organizational users. This shift has enabled non-developers to perform tasks previously bottlenecked by technical expertise, such as data transformation, automation, and debugging. Notably, over 25% of the work performed by business-function employees using these tools is now categorized as engineering or coding, demonstrating that agents lower the barrier to cross-functional execution and allow workers to expand their output beyond traditional job descriptions.