Enterprise Signals: What Frontier Firms Are Doing Differently
OpenAI's Enterprise Signals: What Frontier Firms Are Doing Differently (August 2026; enterprise usage data) found that as of June 2026 its Codex agent generated 64% of combined Codex and ChatGPT output tokens among enterprise customers.
Key findings
- 01As of June 2026, Codex generated 64% of combined Codex and ChatGPT output tokens among OpenAI's enterprise customers.OpenAI uses Codex tokens as its definition of agentic use. Agentic tasks run longer and use more computation, which partly explains the share.
- 02Frontier firms, the top 10% by usage each month, generated 8.3x as many output tokens per active user as typical firms in June, up from 2.6x in January.Typical firms are those between the 45th and 55th percentiles. The gap was largest in information and technology (11.7x) and smallest in manufacturing (5.3x).
- 03Among weekly active users, 21% at frontier firms use plugins and 19% use skills, compared with 9% and 3% at typical firms.Plugins and skills give agents company context and connections to tools such as a CRM. At OpenAI itself, 95% of employees use plugins weekly.
- 04Since February 2026, weekly active enterprise Codex users grew 108x in legal, 41x in sales, 41x in recruiting and 26x in marketing, compared with 5x among engineers.The page reports growth multiples, not absolute user counts, so the size of these groups is not known.
- 05Coding and system or agent operations together account for nearly 75% of agentic messages, while writing remains the most common use of ChatGPT.In recruiting (32%), sales (26%), policy (25%) and communications (24%), a quarter to a third of messages go to system and agent operations.
By the numbers
64%of enterprise output tokens from the Codex agent (June 2026)
8.3xtokens per active user, frontier vs. typical firms (was 2.6x in January)
108xgrowth in weekly active Codex users in legal since February
What it means for you Draft
For executives at $10–100M companies
This is measured usage, not a survey, and it suggests the heaviest AI users are pulling away from typical companies mainly by handing agents longer, multi-step work with access to company tools. Token volume is not business value, as OpenAI itself notes, so treat this as a sign of how deeply AI is being used, not proof of returns. For a $10–100M firm, the practical question may be whether any team has moved from asking AI questions to delegating whole tasks with proper access controls.
For practitioners
The gap between frontier and typical firms shows up in connecting agents to company data and tools (plugins, skills), not in model access. Growth in legal, sales and recruiting suggests non-engineering teams can use agent tools, but the page gives growth multiples, not absolute user counts.
Limitations
Medium trust.Large first-party usage dataset with clear definitions; OpenAI sells the products measured and discloses no sample counts.
OpenAI's own customer data, reported by a vendor that benefits from higher usage; no firm counts or absolute user numbers are disclosed. Tokens are an imperfect proxy for value, and 'frontier' firms are defined by usage itself, so the gap partly measures the ranking.
About the publisher: Makes ChatGPT; studies its own customers.
OpenAI. "Enterprise Signals: What Frontier Firms Are Doing Differently." August 12, 2026. https://openai.com/signals/enterprise-data/
Related reports
Technology Report 2026Leaders expect a 95% developer productivity uplift in 1–2 years; current gains are 20–27%. Bain says AI moves bottlenecks into review and coordination.
Gartner Survey Finds Only 22% of Organizations Have Successfully Scaled AI Across Multiple Business Units (press release)85% of functional leaders plan to increase AI spending in 2026 after allocating an average 12% of functional budgets to AI in 2025.
The state of AI in 2026: On the road to ROI80% say AI improved their individual productivity, yet only about 6% are AI high performers, flat since 2025.