The Shift from Assistance to Execution

Enterprise AI adoption is evolving from simple question-answering to agentic workflows that perform multi-step tasks. While assistants help users think, agents complete work by interacting with tools, creating files, and drafting outputs for human review. This transition is quantified by the rise of Codex usage, which now accounts for 64% of combined output tokens among enterprise customers, reflecting a move toward longer, more complex task completion.

The Widening 'Frontier Gap'

A significant performance gap has emerged between 'frontier firms' (the top 10% of AI users) and typical organizations. As of June, frontier firms generated 8.3x more output tokens per active user than typical firms—a substantial increase from the 2.6x gap observed in January. This disparity is not limited to tech companies; it spans various industries and company sizes. The data suggests that access to models is insufficient; scaling AI requires complementary investments in data infrastructure, employee training, and governance.

Scaling Agentic Workflows

Frontier firms differentiate themselves by integrating advanced capabilities that connect agents to company context. Specifically, these firms are significantly more likely to use:

  • Plugins: Bundled capabilities that connect agents to internal apps and data (e.g., CRMs or proprietary playbooks).
  • Skills: Reusable instructions that standardize how agents approach specific tasks.

While software engineering was the initial driver of this adoption, agentic workflows are spreading rapidly into other knowledge-work functions. Since February, weekly active Codex users grew 108x in legal, 41x in sales, 41x in recruiting, and 26x in marketing, compared to 5x in engineering.

Identifying and Scaling Internal Expertise

Contrary to the assumption that AI adoption is top-down, administrative data shows that early-career employees are more active users than executives, sending 13 more messages per week six months after adoption. Leaders can close the frontier gap by identifying these high-frequency users, making their effective workflows visible, and codifying them into repeatable, organization-wide practices. Success depends on balancing this autonomy with clear governance, permissions, and human review processes.