OpenAI recently published two studies that quantify a shift in enterprise AI: from answering questions to carrying out work. The headline number is stark—frontier firms (top 10% by usage) now generate 8.3× as many output tokens per active user as typical firms, up from 2.6× in January. That gap is a proxy for depth of use, and it appears across industries and company sizes. For product builders and AI practitioners, the data offers a practical roadmap: connect agents to company context and tools, establish governance and review, and turn individual workflows into shared practices.
What Changed: From Asking to Doing
Enterprise AI is moving from assistance to execution. Assistants help people think through work; agents help them complete it. Products like ChatGPT Work and Codex can use tools, create files, and produce work for review. Instead of asking AI how to prepare a presentation, a worker can ask an agent to gather relevant information across sources and draft the presentation itself.
This shift is visible in usage data. As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers. Agentic workflows typically generate more output because they carry out longer, multi-step tasks, so the figure reflects both how often Codex is used and how much output those tasks produce.
For product builders, this signals a change in user expectations. When AI moves from assistance to execution, users stop asking for answers and start asking for deliverables. Your product needs to support agents that can operate tools, access company data, and produce work that humans can review.
The Frontier Gap Widens
OpenAI ranks enterprise customers each month by output tokens per active user. Frontier firms are those in the top 10% that month; typical firms fall between the 45th and 55th percentiles. As of June, frontier firms generated 8.3× as many output tokens per active user as typical firms, a threefold increase from the 2.6× gap in January. The gap appears across industries and company sizes, showing that intensive AI use is not limited to technology companies.
The companion working paper, How Organizations Use AI: Evidence from ChatGPT, finds that among U.S. public companies, enterprise adopters had stronger financial measures compared to non-adopters—more assets, more workers, and higher R&D investment. Together, the reports suggest that access alone may not be enough to scale AI. Complementary investments in continuous employee learning, shared workflows, data infrastructure, and governance support broader and deeper adoption.
For product teams, this means that feature completeness is not the same as adoption depth. You need to help customers build habits, not just provide tools. Consider how your product can encourage regular use, shared workflows, and integration with existing data and governance structures.
Advanced Capabilities: Plugins and Skills
AI agents need access to the right context and tools to be effective. Plugins bundle capabilities that help agents complete specific workflows. They combine skills—reusable instructions—with apps that connect to company data, tools, and actions. For example, a sales Plugin can combine a team’s playbook with access to its CRM, allowing an agent to use current customer information and past proposals to prepare a tailored response for review.
Frontier firms have a clear lead in advanced capabilities. Among weekly active users, 21% at frontier firms use Plugins and 19% use skills, compared with just 9% and 3% at typical firms. However, frontier firm adoption represents only a fraction of what is possible. OpenAI’s internal usage highlights the potential for deeper usage, with weekly Plugin usage at 95% of active users.
This design—bundling skills and apps—is a model for product builders. The combination of context and tools is what enables agents to actually execute. When designing your product, think about how users can package reusable instructions with access to their specific data and systems.
Agents Spread Across Knowledge Work
Software engineering was an early center of agentic adoption, but Codex use is now growing quickly across knowledge-work functions. Since February, weekly active enterprise Codex users grew 108× in legal, 41× in sales, 41× in recruiting, and 26× in marketing, compared with 5× in engineering.
At Virgin Atlantic, that shift is visible across the business. Engineering teams use Codex to refactor legacy code in 30 minutes instead of two weeks. Their product teams use ChatGPT Work to complete weeks of competitive research in hours, shaping the airline’s five-year digital strategy.
For product managers, this is a signal: don’t position agents only as “engineer’s assistants.” Legal, sales, recruiting, and marketing tasks often rely heavily on internal knowledge and tool integrations—exactly where agents can deliver value. Build for these functions, not just for developers.
Early-Career Employees Lead Usage
Many surveys have reported higher levels of AI use among leaders and executives. However, administrative data from millions of conversations finds the opposite. Six months after adoption, early-career employees sent 13 more messages per week than executives.
For leaders, this points to a practical opportunity: identify employees with the strongest AI habits and make their workflows visible, helping effective practices spread across all levels. For product builders, this suggests that your power users may not be the ones you expect. Design for early-career workers who are eager to adopt, and provide ways for their workflows to be shared and scaled.
What Leaders Can Do to Narrow the Gap
Companies may have access to the same models, but frontier firms are putting them to work faster and more deeply across their organizations. The opportunity for leaders is to close the frontier gap by extending agentic workflows beyond engineering and turning successful individual use cases into repeatable practices across the organization.
Concretely, this means: connect agents to company context and tools, with clear permissions, governance, and human review. Then, take the workflows that work for individuals and make them standard practice for teams. This is not a slogan—it is the path to narrowing the frontier gap.
For product builders, the takeaway is clear: the value of AI is not in the model itself but in how organizations make it part of daily execution. The next step is not to add more features but to help customers move from occasional use to weekly habits, and to enable agents to actually complete work. OpenAI’s full report, Enterprise Signals, provides more industry and function details for those who want to dive deeper.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.