From assistance to execution: How enterprises put AI to work

OpenAI published two studies examining how enterprises are adopting AI agents that perform tasks rather than just provide answers. The research highlights a widening performance gap between 'frontier firms'—those in the top 10% of AI usage—and typical companies. Frontier firms now generate 8.3 times as many output tokens per active user as typical firms, up from a 2.6 times gap in January, indicating deeper and more productive AI integration across industries and company sizes.

The reports emphasize that simply adopting AI tools is insufficient; organizations must invest in employee training, shared workflows, and governance to scale usage effectively. Enterprise Signals and the companion paper 'How Organizations Use AI: Evidence from ChatGPT' analyze adoption patterns across roles and seniority levels, noting that early-career employees often lead in AI message volume six months after adoption, challenging assumptions about leadership-driven AI use.

Agentic AI capabilities, such as plugins and skills, are enabling tasks like drafting presentations, preparing customer responses, and refactoring code. Codex, for example, generated 64% of combined Codex and ChatGPT output tokens among enterprise customers as of June, reflecting its role in executing multi-step workflows. Frontier firms are far more likely to use advanced features like plugins (21%) and skills (19%) compared to typical firms (9% and 3%, respectively).

The studies suggest that companies must connect AI agents to company-specific context and tools to maximize effectiveness. Virgin Atlantic’s engineering teams reportedly refactored legacy code in 30 minutes instead of two weeks, while product teams completed weeks of competitive research in hours. OpenAI’s internal data shows potential for even deeper adoption, with 95% of weekly active users leveraging plugins, underscoring the opportunity for organizations to close the frontier gap by standardizing successful AI workflows.