{"slug": "openai-data-finds-top-enterprise-users-adopt-reusable-ai-skills-six-times-as", "title": "OpenAI data finds top enterprise users adopt reusable AI skills six times as often", "summary": "OpenAI, led by CEO Sam Altman, published two studies on August 12 showing that its top enterprise users adopt reusable AI skills roughly six times as often as typical customers and use Plugins at over twice the rate. The reports, based on OpenAI's own customer telemetry, indicate that frontier firms generated 8.3 times as many output tokens per active user as typical firms in June, up from a 2.6-times gap in January.", "body_md": "[OpenAI](https://openai.com/?ref=runtimewire), led by co-founder and CEO [Sam Altman (@sama)](https://x.com/sama?ref=runtimewire), published two studies on August 12 arguing that the enterprise AI divide is increasingly determined by how companies organize work around agents. Its heaviest users adopt reusable skills roughly six times as often as typical customers and use Plugins at over twice the rate, according to [OpenAI's announcement](https://openai.com/index/how-enterprises-put-ai-to-work?ref=runtimewire).\n\n[OpenAI on X](https://x.com/OpenAI/status/2087912623883051300?ref=runtimewire)\n\nFor founders and independent operators, the reports offer a deployment playbook with an important limitation: this is OpenAI's analysis of its own customer telemetry, rather than an independent or cross-vendor study of enterprise AI.\n\n[Altman studied computer science at Stanford for two years before leaving to co-found Loopt](https://ecorner.stanford.edu/wp-content/uploads/sites/2/2024/05/the-possibilities-of-ai-entire-talk-transcript.pdf?ref=runtimewire), a location-based social-networking company. He later went from founder to seed investor and president of Y Combinator. In a [2018 YC interview](https://www.ycombinator.com/blog/uncut-interview-with-sam-altman-on-masters-of-scale/?ref=runtimewire), he described missing the intensity of operating while he was investing from the sidelines. OpenAI's latest enterprise research carries that operator's focus: access to a powerful model is treated as the starting condition, while repeatable deployment is the work.\n\nThat framing also serves OpenAI's commercial strategy. The reports present ChatGPT Work, Codex, Plugins, skills and connected apps as pieces of an operating system for enterprise agents. OpenAI wants customers to move past isolated conversations and give agents the instructions, company data and tools required to complete multistep assignments.\n\n### What the usage gap measures\n\nOpenAI's [Enterprise Signals report](https://openai.com/signals/enterprise-data/?ref=runtimewire) divides customers into two groups. \"Frontier firms\" are the top 10% each month by output tokens generated per active user. \"Typical firms\" sit between the 45th and 55th percentiles.\n\nBy that measure, frontier firms generated 8.3 times as many output tokens per active user as typical firms in June, up from a 2.6-times gap in January. The result shows that heavy users are pulling further away on the metric used to identify them. It does not establish that they are producing better work or earning higher returns.\n\nOpenAI acknowledges that limitation. A long response can have little value, while a short response can settle an expensive question. Token volume is a proxy for how much work users hand to AI, and longer agent assignments naturally produce more tokens than short chat exchanges.\n\nThe capability data adds a more useful distinction. Among weekly active users, 21% at frontier firms used Plugins, compared with 9% at typical firms. Skills, which package reusable instructions for recurring work, reached 19% at frontier firms and 3% at typical firms. OpenAI says 95% of its own active employees use Plugins each week, an internal comparison that doubles as a target for customers.\n\nA sales Plugin, for example, can combine a company's playbook with access to its customer relationship management system. An agent can then retrieve account history, consult previous proposals and prepare a tailored response for a person to review. The workflow moves useful knowledge out of an employee's private prompt history and into a shared process.\n\nThat is OpenAI's central product bet. The durable enterprise layer will be built from permissions, integrations, reusable instructions and review systems around the model. OpenAI is using its own customer telemetry to tell executives what to buy and deployment teams what to build.\n\n### Codex moves outside engineering\n\nCodex generated 64% of the combined Codex and ChatGPT output tokens among OpenAI's enterprise customers in June. That figure reflects both adoption and the greater volume produced by agents completing longer tasks, so it cannot be read as a simple share of users or assignments.\n\nThe direction is still clear. Since February, OpenAI says weekly active enterprise Codex users increased 108 times in legal, 41 times in sales, 41 times in recruiting and 26 times in marketing. Engineering grew five times from a more established base.\n\nOpenAI's broader [Enterprise Signals analysis](https://openai.com/signals/enterprise-data/?ref=runtimewire#L91-L94) draws on over 10 million messages and finds that coding plus system and agent operations account for nearly three-quarters of agentic messages. System and agent operations represent 32% of agentic messages in recruiting, 26% in sales, 25% in policy and 24% in communications.\n\nThe adoption pattern supports Altman's push to distribute Codex beyond a developer-only product. Enterprise growth increasingly depends on putting the same execution tools in front of lawyers, recruiters, marketers and finance teams.\n\n### A workforce study with a different lens\n\nOpenAI published a companion working paper, [\"How Organizations Use AI: Evidence from ChatGPT\"](https://arxiv.org/html/2608.12236?ref=runtimewire), examining account records, worker roles, message classifications and public-company financial data through March 2026.\n\nThe paper's [worker-characteristics sample](https://arxiv.org/html/2608.12236?ref=runtimewire#L100-L114) covers 1,764 organizations and 17,446,551 messages. The researchers used de-identified data, mapped job titles to broad categories and classified message content automatically; they did not manually review individual messages. At six months after adoption, [early-career workers and trainees sent roughly eight to nine more messages per week](https://arxiv.org/html/2608.12236?ref=runtimewire#L181-L187) than the average active user within the same firm. Executives sent fewer messages, but the paper does not report a 13-message gap between the groups.\n\nThe paper also found that [US public companies adopting ChatGPT Enterprise tended to be larger, employ more people and spend more on research and development](https://arxiv.org/html/2608.12236?ref=runtimewire#L115-L120) than companies without a ChatGPT Enterprise ticker match. The comparison group should not be treated as companies using no AI, a limitation [the researchers acknowledge](https://arxiv.org/html/2608.12236?ref=runtimewire#L157-L165). Those characteristics complicate any claim that AI access itself produces stronger companies. Organizations already equipped with capital, technical staff and management capacity may have an easier time adopting AI deeply.\n\nEnterprise Signals has a related constraint. OpenAI analyzes customers already paying for its products, making the results a study of adoption inside OpenAI's customer base rather than a benchmark for the full enterprise market. The report's aggregated methodology also leaves the size and composition of the frontier and typical customer cohorts unspecified.\n\n### OpenAI's enterprise playbook\n\nThe reports give operators a practical sequence: find employees already building effective workflows, connect agents to approved data and tools, define where people must review decisions, and convert successful experiments into shared systems. Governance becomes part of deployment because agents with access to customer records, financial systems or internal communications require explicit boundaries and auditable actions.\n\nFor startup teams, the useful lesson is operational rather than comparative. Shared instructions, controlled access and review procedures can turn an effective employee experiment into a repeatable company process, but OpenAI's data does not show that these practices improve revenue or productivity.\n\nFor OpenAI, the research supports a larger shift from selling model access to owning the layer where companies configure and supervise AI work. OpenAI's March 31 financing brought in [$122 billion at an $852 billion post-money valuation](https://openai.com/index/accelerating-the-next-phase-ai/?ref=runtimewire), according to OpenAI. That capital raised the commercial stakes around enterprise distribution and recurring use.\n\nThe strongest result in the new data is the gap in organizational behavior. Companies buying access to the same models are producing sharply different usage patterns. The heavy users package knowledge into skills, connect agents to operational systems and spread working methods across teams. Whether those habits create corresponding gains in revenue, quality or labor productivity remains outside the studies' evidence. OpenAI has shown where usage compounds. Customers still have to prove where value does.", "url": "https://wpnews.pro/news/openai-data-finds-top-enterprise-users-adopt-reusable-ai-skills-six-times-as", "canonical_source": "https://runtimewire.com/article/openai-enterprise-ai-skills-plugins-frontier-firms", "published_at": "2026-08-13 14:52:44+00:00", "updated_at": "2026-08-13 15:07:33.471251+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-agents", "ai-research"], "entities": ["OpenAI", "Sam Altman", "ChatGPT Work", "Codex", "Plugins", "Enterprise Signals"], "alternates": {"html": "https://wpnews.pro/news/openai-data-finds-top-enterprise-users-adopt-reusable-ai-skills-six-times-as", "markdown": "https://wpnews.pro/news/openai-data-finds-top-enterprise-users-adopt-reusable-ai-skills-six-times-as.md", "text": "https://wpnews.pro/news/openai-data-finds-top-enterprise-users-adopt-reusable-ai-skills-six-times-as.txt", "jsonld": "https://wpnews.pro/news/openai-data-finds-top-enterprise-users-adopt-reusable-ai-skills-six-times-as.jsonld"}}