Two new reports from OpenAI describe how businesses are using artificial intelligence. The reports look at OpenAI's enterprise customers and how their use of the technology has changed over time.
The main finding is a shift in how workers use AI. Instead of asking a chatbot for advice, workers are now asking AI tools to complete full tasks on their own.
OpenAI calls this a move "from assistance to execution." Assistants answer questions. Agents finish jobs like drafting documents or writing code.
The Gap Between Top Firms And Average Firms
The reports split companies into two groups. Frontier firms are the top 10% of users each month. Typical firms fall in the middle of the pack.
As of June, frontier firms produced 8.3 times more AI output per active user than typical firms. In January, that gap was only 2.6 times.
This gap shows up across many industries and company sizes. It is not limited to technology companies.
Companies that adopted AI early also tend to have stronger finances. The report found these firms held more assets, employed more workers, and spent more on research.
Codex, OpenAI's coding and task tool, made up 64% of all enterprise output tokens as of June. This is combined with ChatGPT usage. Output tokens are a way to measure how much work an AI tool produces.
Agentic tools tend to produce more output because they complete longer tasks with several steps. That helps explain why Codex use has grown so fast.
Advanced Tools Spread Faster At Top Firms
Frontier firms use more advanced AI features than typical firms. These features are called Plugins and skills.
Plugins combine instructions with access to company data and tools. An example given in the report is a sales plugin that connects a team's playbook to its customer database.
At frontier firms, 21% of weekly active users use Plugins. At typical firms, that number drops to 9%. Skills show a similar gap, at 19% versus 3%.
Inside OpenAI itself, 95% of employees use Plugins each week. The company says this shows how much room there still is for growth at other firms.
Codex use is also spreading beyond engineering teams. Since February, weekly Codex users grew 108 times in legal work, 41 times in sales, 41 times in recruiting, and 26 times in marketing.
Engineering, by comparison, grew 5 times over the same period. This suggests agentic AI tools are being adopted by workers outside of technical roles.
Virgin Atlantic is named as one example. The airline's engineering team says Codex helps them update old code in 30 minutes instead of two weeks.
The airline's product team also uses ChatGPT Work to speed up research that once took weeks. That research now takes hours, according to the report.
One more finding stood out. Early-career workers send more AI messages per week than executives, even six months after adoption began.
The report suggests leaders could look at how their most active employees use AI. Sharing those habits across teams may help other workers catch up.
OpenAI says enterprise customers can request a custom report comparing their own AI use to frontier firms. Both full reports are available for the public to read on OpenAI's site.
