OpenAI pulls back the curtain on enterprise AI

If you thought giving employees an AI subscription would instantly cut the standard 40-hour workweek in half, OpenAI’s latest internal numbers bring a reality check.

OpenAI dropped its "The State of Enterprise AI" report, pulling together anonymized data from over 1 million business customers and surveying 9,000 workers across nearly 100 organizations. With ChatGPT now counting over 800 million weekly active users, OpenAI wanted to see how that massive consumer surge translates to actual corporate desk work.

The headline takeaway? AI is everywhere in the office, but for the average worker, it is currently saving only 40 to 60 minutes per day.

Complex tokens soar, but routine time savings stay modest

By raw adoption metrics, the expansion looks staggering. Over the past year, weekly messages in ChatGPT Enterprise jumped roughly 8x, and an average employee sends about 30% more prompts. Structured setups like Projects and Custom GPTs skyrocketed 19x year-to-date, signaling a move from casual Q&A into repeat workflows.

Even more dramatic: average reasoning token consumption per enterprise surged roughly 320x over the last 12 months. Tech, healthcare, and manufacturing led adoption growth, while professional services and finance deployed at the largest scale. Outside the U.S., corporate API customer growth topped 70% in six months, with Japan hosting the biggest corporate API base.

Workers clearly appreciate the assist. Three out of four respondents said AI boosted output speed or quality, and 75% said they could tackle tasks they previously could not do. Non-technical staff sent 36% more coding-related prompts, and 87% of IT workers reported faster troubleshooting.

Yet as outlets like CNET pointed out, cutting 40 to 60 minutes out of a packed calendar of meetings, Slack pings, and email threads hardly amounts to the radical productivity transformation enterprise pitches have promised.

The widening gap between casuals and power users

The report reveals a stark internal divide. The top 5% of users—what OpenAI classifies as "frontier workers"—are pulling away from the median pack.

These heavy users send 6x more messages than the median employee and deeply integrate advanced tool sets into their daily routines. For them, time saved exceeds 10 hours a week—roughly two hours a workday. Similarly, "frontier firms" see double the messages per seat compared to typical companies.

OpenAI notes that it rolls out a new feature or model capability roughly every three days. The bottleneck to getting more out of AI is no longer raw intelligence or API tooling; it is how fast corporate workflows and traditional management structures can adapt.

Why it matters

We think this report provides a much-needed grounding for the enterprise AI conversation. AI is undeniably helpful, but right now it behaves more like an ultra-responsive intern than an autonomous colleague. For rank-and-file workers using it as a faster search engine or copy polisher, an hour a day might well be the natural ceiling.

For enterprise tech vendors, the 320x surge in reasoning token usage paired with modest daily time savings highlights a critical friction point: models are getting dramatically smarter, but corporate processes are still manual. CIOs will increasingly struggle to justify steep seat licenses if the actual net return stays under an hour per employee per day.

In our view, the dividing line over the next year will not be which frontier model a company licenses, but whether workers learn to operate like the top 5%. If organizations do not actively restructure roles around integrated agents and Custom GPTs, AI will remain a convenient sidekick rather than the massive business engine tech executives keep selling.