OpenAI Enterprise AI Adoption: From Assistance to Execution

Enterprise AI is Shifting from Assistance to Execution

Enterprise AI adoption is transitioning from a model of answering questions (assistance) to carrying out substantive work (execution). While assistants help users think through tasks, agents—powered by tools like ChatGPT Work and Codex—enable the completion of multi-step workflows, such as gathering information across sources to draft a full presentation rather than simply advising on how to prepare one.

This shift is quantified by the volume of output tokens. As of June, Codex generated 64% of the combined output tokens for Codex and ChatGPT among enterprise customers, indicating a move toward delegated, agentic work that typically produces more output due to the complexity of the tasks performed.

The Widening "Frontier Gap" in AI Usage

There is a growing disparity in how organizations utilize AI, termed the "frontier gap." Frontier firms—defined as those in the top 10% of AI usage per month—now generate 8.3Ñ as many output tokens per active user as typical firms (those in the 45th to 55th percentiles). This gap has widened significantly from 2.6Ñ in January.

Key findings regarding this gap include:

Cross-Industry Prevalence: The frontier gap exists across various industries and company sizes, not just within technology firms.
Financial Correlation: Data from U.S. public companies indicates that enterprise adopters generally possess stronger financial measures, including more assets, more employees, and higher R&D investment than non-adopters.
Scaling Requirements: Access to models is insufficient for scaling; deeper adoption requires complementary investments in data infrastructure, governance, continuous employee learning, and shared workflows.

Adoption of Advanced Capabilities and Plugins

Frontier firms leverage advanced capabilities—specifically Plugins and skills—at a significantly higher rate than typical firms to connect agents to company context and tools.

Plugin Usage: 21% of weekly active users at frontier firms use Plugins, compared to 9% at typical firms.
Skill Usage: 19% of weekly active users at frontier firms use skills, compared to 3% at typical firms.
Internal Benchmark: At OpenAI, 95% of employees use Plugins weekly, suggesting a high ceiling for potential enterprise adoption.

Plugins bundle skills (reusable instructions) with apps (connections to company data and tools), allowing agents to perform specific workflows, such as a sales agent using a CRM and a team playbook to draft tailored customer responses.

Expansion of Agentic AI Across Knowledge Work

While software engineering was the initial primary driver of agentic AI, Codex usage is now expanding rapidly into other knowledge-work functions. Since February, weekly active enterprise Codex users have grown by the following magnitudes:

Legal: 108Ñ
Sales: 41Ñ
Recruiting: 41Ñ
Marketing: 26Ñ
Engineering: 5Ñ

Real-world application is exemplified by Virgin Atlantic, where engineering teams use Codex to refactor legacy code in 30 minutes instead of two weeks, and product teams use ChatGPT Work to condense weeks of competitive research into hours to inform a five-year digital strategy.

AI Usage Patterns by Employee Seniority

Administrative data from millions of conversations reveals that AI adoption is highest among early-career employees and decreases as seniority increases. Six months after adoption, early-career employees sent an average of 13 more messages per week than executives.

This trend suggests a comparative advantage for early-career workers and provides an opportunity for leadership to identify high-performing AI users and institutionalize their individual workflows as shared organizational practices.