Agents expand spend beyond seats

OpenAI is turning business subscriptions into a usage ramp, not just a seat business. Once Codex and ChatGPT Work are wired into email, calendar, docs, code, and internal tools, each extra agent run creates more model usage and more paid credits, even if the number of employees on the plan stays flat. That matters because agent workflows expand spend through actual work completed, not headcount growth alone.1 2 3

• In practice, Codex is already being used as a daily work layer, reading email and calendars, drafting replies, moving events, scanning customer calls, and running about twenty recurring automations for one power user. That is the kind of behavior that drives incremental usage far beyond a basic chat seat.4
• The same pattern shows up in teams. At Scale AI, agents are used for QC reviews, dashboards, social scheduling, taxonomy changes, and computer use evals. The workflow often starts with a file or prompt, then fans out across Slack, spreadsheets, internal tools, and code, which makes spend scale with task volume.5
• This is also the competitive answer to Claude Code and other agent products. Anthropic showed that a high engagement coding client can become a major revenue stream on its own, and OpenAI is extending that logic beyond coding into broader white collar workflows, where every additional delegated task can meter more usage.2 6

The next step is a fuller hybrid model where companies buy seats for access and governance, then expand spend through credits as more teams trust agents with real work. If OpenAI keeps pushing Codex and ChatGPT Work deeper into everyday workflows, revenue growth should increasingly follow agent activity inside each customer, not just new seat sales.1 3 4 5