The Agentic Usage Gap in Enterprise AI

The Shift to Agentic Delegation

OpenAI research indicates that the gap between frontier firms (top 10% of usage) and average enterprises has widened from 2.6x to 8.3x in output tokens per active user between January and June. This divergence is driven by a fundamental shift from chat-based interaction to agentic delegation. While chat usage remains valuable for drafting and search, agentic workflows allow for systems-level work that impacts broader organizational processes. By June, agentic output tokens accounted for 64% of total enterprise volume, up from 13% in February.

Adoption Patterns in Frontier Firms

Frontier firms differentiate themselves through higher utilization of plugins and skills. Plugins provide connectivity to external applications and data, while skills offer reusable instructions for common workflows. At typical firms, only 9% of weekly active users utilize plugins and 3% use skills. In contrast, frontier firms see 21% plugin adoption and 19% skill adoption. Non-technical departments are seeing the fastest growth in agentic tool usage, with legal departments showing 108x growth in Codex usage since February, followed by sales and account management at 41x, and marketing at 26x.

Overcoming Institutional Inertia

Despite the availability of advanced tools, individual knowledge workers often default to legacy workflows due to ingrained muscle memory. The transition to agentic AI requires moving beyond simple generation tasks toward delegation, where agents are provided with sufficient context, tools, and verification criteria to complete complex, multi-step objectives. The primary barrier to this transition is not model capability, but the psychological and organizational inertia that keeps users performing rote tasks manually rather than restructuring their work to leverage autonomous agents.