Enterprise signals: What frontier firms are doing differently

Enterprise AI is entering a new phase: more work is being delegated to agents, and the firms furthest along are pulling away. Frontier firms have access to the same AI models as other enterprises, but they are deepening their use faster. What sets them apart is how they put those models to work: moving from assistance to delegation, giving agents the context and tools to complete complex tasks, and accelerating agentic AI beyond software development.

Our research on how agents are transforming work at OpenAI offers a leading indicator of what this shift can look like inside an AI-native organization. Enterprise Signals widens the lens to OpenAI’s global enterprise customer base, showing how agentic work is spreading across organizations and what separates frontier firms from the rest.

In this edition of Enterprise Signals, we show that frontier firms—those in the top 10% of AI usage each month–are widening the gap from typical firms–those in the middle 10%. We examine what frontier firms are doing differently and what organizations can do to close the gap.

Key takeaways:

• Enterprise use is becoming more agentic. As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers, suggesting that agents are enabling a shift toward more substantive, delegated work.
• The frontier gap is widening. Frontier firms—those in the top 10% of AI usage each month—now generate 8.3× as many output tokens per active user as typical firms, up from 2.6× in January.
• Frontier firms use advanced capabilities more often. Each week, 21% of active users at frontier firms use Plugins, compared with 9% at typical firms. At OpenAI, 95% of employees use Plugins weekly, highlighting the potential for deeper adoption.
• Agents are spreading across knowledge work. 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.
• Early-career employees use AI more: Usage is highest among early-career workers and falls among more senior employees, suggesting a potential comparative advantage in using AI.

Enterprise AI is moving from answering questions to carrying out work. This shift first took hold in software engineering with Codex. Now, ChatGPT Work is extending agentic AI beyond developers, enabling workers across the enterprise to move from asking for help to delegating substantive tasks.

As of June 2026, agentic AI use (defined as Codex tokens) accounted for 64% of combined Codex and ChatGPT output tokens among enterprise customers. Both ChatGPT and agentic AI use the same underlying frontier models; the difference is how that intelligence is put to work.

ChatGPT helps people ask questions, develop ideas, and work through problems. Agentic AI (like Codex and ChatGPT Work) gives the model tools to work with your computer, allowing it to find information, edit files, and carry out multi-step tasks autonomously or under supervision. Completing complex, long-running tasks can require substantially more computation, helping explain why agentic token usage has grown rapidly.

Each month, we rank enterprise customers by output tokens per active user. Frontier firms are those in the top 10% that month, while 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 compared to the 2.6× gap in January.

The frontier gap also extends across industries, with the largest token usage gap in information and technology (11.7×) and the smallest in manufacturing (5.3×). In contrast, token usage among typical firms is similar across industries, with relatively modest growth over the past year (1.9× to 2.8×). This suggests many organizations are still using simple chat assistants, leaving meaningful room to deepen adoption and  utilize agents to complete more complex work.

Tokens are an imperfect measure of business value: a short response can be highly valuable, while a long one may add little. But token volume offers a useful proxy for the depth of AI use and how much work employees are asking AI to take on.

Leading firms are deepening their use of AI because they are applying it to concrete problems inside their organizations and across their industries.

Early enterprise AI applications started internally, helping employees find information, draft documents, analyze data, write code, and automate routine processes. Leading firms are increasingly extending AI beyond internal productivity and into customer-facing products, services, and experiences.

AI agents need access to the right context and tools to be effective. Plugins⁠(opens in a new window) bundle capabilities that help agents complete specific workflows. They can combine skills that provide 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 of these capabilities, with weekly plugin usage at 95% of active users. Our research on how agents are transforming work provides a closer look at how employees at OpenAI use advanced capabilities.

AI agents need three things to complete meaningful work: context, tools, and persistence. Memory, voice input, and appshots help workers to easily share the context and information necessary to successfully complete a task. Computer and browser use allow agents to navigate websites, create files, and complete tasks autonomously or under supervision. Goals and loops keep agents working until a task is finished. Together, these capabilities help frontier firms realize the potential of agents.

Giving agents access to company systems also introduces new risks. Frontier firms set clear rules for where agents can operate, what information they can access, when they can take actions, and how people review higher-risk decisions. These controls must evolve as organizations learn from real deployments.

Frontier firms also make continuous learning and experimentation part of everyday work. Hands-on builder sessions, forums for sharing emerging use cases, and internal showcases help successful practices spread across teams and functions.

Software developers were among the first to adopt AI agents, but the fastest growth is now coming from general knowledge workers. Since February, the number of weekly active enterprise Codex users has grown 108× in legal, 41× in sales, 41× in recruiting, and 26× in marketing, compared with 5x among engineers. 

Software moved first for a reason. Codebases give agents clear context, tests make outputs easier to verify, and progress in coding helps accelerate AI research and development. Over the past two years, developers have moved from using AI to complete individual lines of code to having agents handle entire software development tasks. Many developers are no longer writing code by hand, instead focusing their time on setting up tasks, orchestrating agents, and reviewing results.

In contrast, progress in general knowledge work has been slower because many tasks provide limited context, can be difficult to specify, and lack clear criteria for verifying the result. But continued scaling, advances in reinforcement learning, and targeted efforts to improve performance on evaluations such as GDPval are bringing more real-world tasks, tools, and work environments within reach of frontier models. As a result, agentic AI has increasingly found product-market fit with general knowledge workers since the beginning of the year.

Across a sample of more than 10 million messages, enterprise AI is moving from helping employees produce work to helping teams carry it out. Even within the same function, usage patterns differ significantly between chat and agentic AI.

Writing remains the most common use of ChatGPT, while coding and system or agent operations together account for nearly 75% of agentic messages. This shift is especially pronounced outside traditional technical teams: recruiting (32%), sales (26%), policy (25%), and communications (24%) devote between a quarter and a third of messages to system and agent operations. Technical work is also spreading across functions, with coding accounting for almost 60% of agentic messages in design.

There is no single AI adoption leaderboard. Industry rankings vary depending on the measure used. In the past month, Professional and Scientific Services led both Codex adoption and API intensity. Arts, Entertainment, and Recreation led ChatGPT adoption, while Manufacturing has the highest ChatGPT intensity despite ranking last in Codex and API intensity.

Industries are not following the same path to AI adoption. Some are advancing through developer tools and AI-powered products, while others are seeing broader or more intensive everyday use.

As AI creates new ways for organizations to deliver goods and services, it is also reshaping the security landscape. Few risks are more immediate than cybersecurity, especially given the rapid pace of AI cyber capabilities.

AI can also help defenders respond. Models can analyze large codebases, investigate vulnerabilities, and support multi-step security work. But identifying more issues does not automatically make systems safer. The real bottleneck is turning findings into action: determining which vulnerabilities are genuine, prioritizing the most consequential risks, testing patches, and ensuring that security teams review and deploy fixes quickly. 

OpenAI Daybreak brings together frontier cybersecurity models, enterprise security tools, and a partner ecosystem to help organizations meet these challenges and strengthen their defenses against emerging threats.

OpenAI’s mission is to ensure that AI benefits all of humanity. Organizations will shape much of AI’s economic impact, and for this reason, we regularly publish Enterprise Signals, a recurring set of measures tracking enterprise AI adoption, diffusion, and impact. All analyses in this report are based on aggregated, de-identified enterprise usage data. We used automated systems to classify message content, and no OpenAI employee reviewed customer messages.

OpenAI Enterprise customers can also request a customized benchmark to see how their organization compares across the latest usage and capability measures.