Companies Are Moving from AI Assistants to Delegating Entire Workflows to Agents

According to OpenAI data, enterprise use of artificial intelligence is shifting from individual queries toward delegating multi-step work. On August 12, the company published two studies based on usage of its enterprise services. It defines “frontier firms” as the ten percent of customers with the highest monthly volume of output tokens per active user. In June, these customers generated 8.3 times more output per user than a typical company. In January, the gap was only 2.6 times. The figure itself does not show that higher consumption automatically means higher productivity. It does, however, show a difference in deployment depth. The most active organizations do not use AI merely as a general conversational tool. They are more likely to connect it to company data, tools, and predefined workflows. Each week, 21 percent of active users at the most advanced firms use these expanded capabilities, compared with nine percent at typical customers. A marked change is visible in the Codex programming agent. In June, Codex accounted for 64 percent of the combined output-token volume generated by business customers through Codex and ChatGPT. Since February, weekly Codex users have grown fastest outside IT itself: 108-fold in legal teams, 41-fold in sales and recruiting, and 26-fold in marketing. In technical teams, usage increased fivefold. For management, the implication is significant: buying licenses is only the beginning. A company must determine which data sources an agent may access, which actions it may perform, and when a person must take over responsibility for the output. OpenAI explicitly recommends combining access to internal context with permissions, controls, and governance rules. Another practical management task is turning successful individual practices into repeatable processes for entire teams. The generational dimension of adoption is also notable. OpenAI says employees at the beginning of their careers use AI more often than senior leaders. That could reverse the usual model of technology adoption: some of the most effective practices may emerge not from leadership or IT but from employees who experiment with the tools every day. Companies therefore have reason to identify specific workflows that save time or improve quality instead of merely counting active accounts. OpenAI argues that the main difference between companies is not access to the same models but their ability to integrate those models into real work. This stage will also determine return on investment: costs are shifting away from the license price alone toward process design, integrations, permission management, and output control. Key terms AI agent: A system that can not only generate an answer but also independently carry out several linked steps and use additional tools. Token: A basic unit of text processed by a language model; token volume affects both usage and cost. Workflow: A predefined sequence of steps, data, responsibilities, and controls required to complete a task.

Companies Are Moving from AI Assistants to Delegating Entire Workflows to Agents

The full article is available to users registered at Hard Skills