OpenAI: Enterprise agents now generate 64% of output tokens
OpenAI published data on August 12, 2026, showing that enterprise customers are increasingly deploying agentic AI systems for autonomous work. According to the company's enterprise report "From assistance to execution: How enterprises put AI to work," OpenAI Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers as of June 2026.
Shift from assistance to execution
The metric reflects a fundamental change in how large organizations are using AI. Rather than deploying models primarily for human-in-the-loop assistance—such as code suggestions or writing help—enterprises are now tasking agents with completing work independently. OpenAI stated directly: "Enterprise use is becoming more agentic" and noted that "Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers, suggesting that agents are enabling a shift toward more substantive, delegated work." OpenAI
Codex, OpenAI's code-generation engine, powers autonomous coding agents that can author, test, and deploy software with minimal human oversight. The dominance of Codex output among enterprise customers indicates that organizations are prioritizing task delegation—having agents execute concrete work—over interactive assistance modes.
What the numbers mean
The 64% figure measures token consumption, a proxy for computational effort and system usage. A higher Codex share suggests that enterprises are running more agentic workloads: agents that make decisions, take actions, and produce output without waiting for human review at each step. This contrasts with ChatGPT's primary use case, which remains conversational assistance and human collaboration.
The finding arrives amid broader industry adoption of AI agents across finance, software development, customer service, and logistics. Enterprises are reporting measurable outcomes—reduced deployment times, cost savings, and labor reallocation—though OpenAI's August 2026 report focused on token metrics rather than business results.
Timing and significance
The June 2026 snapshot captures a critical inflection point. Enterprise AI adoption has accelerated over the past 18 months, with organizations moving beyond pilots into production deployments. By measuring token generation rather than API calls or user counts, OpenAI highlighted actual computational consumption—a more granular indicator of where enterprise workloads are concentrating.
The report positions agents as the next frontier of enterprise AI ROI. Rather than asking employees to interact with AI tools, companies are now building systems where AI agents interact with other systems, execute workflows, and handle exceptions automatically. OpenAI's data validates that this transition is already measurable in real usage patterns among its largest customers.