What OpenAI's Data Reveals About the Future of Work

OpenAI just released its first comprehensive research at the State of Enterprise AI adoption. The study is underpinned with reliable primary sources as it surveys 9,000 workers across nearly 100 enterprises.

For Enterprise leaders, a revealing insight emerges: frontier workers, which the study defined as those in the 95th percentile of adoption intensity, send 6x more messages than the median employee and report saving the most amount of hours per week (>10 hours/week).

Worker Productivity Divide Stems from AI Usage Intensity

The 6x message gap between frontier workers (95th percentile) and median workers is just the headline. The task-specific gaps tell the real story.

Coding exhibits a 17x gap. Frontier workers send 17 times more coding-related messages than median peers. Data analysis shows a 16x gap among analysts themselves. These are not marginal differences as they represent fundamentally different relationships with AI as a work tool.

The productivity implications are measurable. Workers who engage across seven task types report 5x more time saved than those using only four task types. Engineering and data science teams save 60-80 minutes per active day.

And the correlation appears to be is direct: depth of use — measured by token consumption and total number of AI tool usage — is leading to higher reported time saved and task completion. That’s because workers are reporting expanded capacity to do tasks that workers report to not have been able to do prior to LLMs. This involves tasks like coding, data analysis and IT trouble shooting.

In total, 75% of users report being able to complete new tasks, and those who save the most time use multiple models, engage with more tools, and use AI across a wider range of tasks.

Yet 19% of monthly active ChatGPT Enterprise users have never touched data analysis tools. 14% have never used reasoning capabilities. Adoption, not access, appears to be a constraint. And this represents an opportunity for enterprise leaders with regards to change management.

From Assistants to Agents: The Shift That Matters

The next phase of enterprise AI represents a fundamental shift. OpenAI's report describes it directly: organizations are moving "from asking models for outputs to delegating complex, multi-step workflows."

In simple terms, this is the difference between asking AI for help and delegating work to AI. That’s because Agents can execute sequences, make decisions, and complete tasks autonomously.

The 320x increase in API reasoning token consumption signals this shift is already underway.

The evidence is already visible in customer experience. Oscar Health answers 58% of benefits questions instantly. These are not pilots. They are production deployments handling millions of interactions.

Multi-agent AI systems are delivering even higher automation rates. At Inkeep for example, our customers are already gains with AI Agents. For instance, Payabli achieves approximately 80% deflection with AI agents for customer support; and Fingerprint reduced tickets by 48% in A/B testing while improving activation by 18%.

The pattern is consistent: sophisticated agent orchestration enables time savings and impact on business outcomes when teams are prepared to leverage it.

External research reinforces the stakes. A 2025 BCG study found that AI leaders, compared to laggards, achieved:

• 1.7x revenue growth
• 3.6x greater total shareholder return
• 1.6x EBIT margin improvement

The gap between frontier worker adoption and median isn't just a productivity metric—it's a competitive advantage metric for enterprises.

Customer service and content generation now represent roughly 20% of API activity. Developers, CX and GTM teams are the proving ground for what comes next across the enterprise.

The Companies Using Agents Are Already Ahead

Here's what the data doesn't say explicitly but implies clearly: the organizations deploying AI agents today aren't just early adopters—they're building durable competitive advantages.

The 6x frontier gap exists because power users have learned to orchestrate AI manually. But manual orchestration doesn't scale. The companies that have moved from "frontier workers using ChatGPT" to "AI agents handling multi-step workflows autonomously" are capturing compounding gains that manual users simply cannot match.

And the window is closing.

At Inkeep, we believe that AI Agent orchestration in 2026 won't be a differentiator as it is today, but rather table stakes. The same way mobile-first strategies separated winners from losers in the 2010s, agent-native architectures will define competitive positioning in the next cycle. Organizations still relying on individual ChatGPT productivity in 18 months will be the ones explaining to boards why they're behind.

The difference between moving now and moving later isn't incremental—it's structural.

What Frontier Organizations Do Differently

The data reveals a consistent pattern among leading enterprises:

1. They integrate AI into systems, not just workflows. Roughly one in four enterprises still hasn't enabled connectors that give AI secure access to company data. Frontier firms treat AI as infrastructure, not a standalone tool.

2. They standardize and share. Usage of Custom GPTs and Projects increased 19x year-to-date. Leading organizations like BBVA regularly use more than 4,000 GPTs—turning individual workflows into reusable, institutional capabilities.

3. They measure depth, not just breadth. Frontier firms send 2x more messages per seat and 7x more messages to GPTs than the median enterprise. They're not just adopting AI more widely; they're embedding it more deeply into core operations.

4. They're already building agent systems. Top API use cases include in-app assistants, customer support, and—notably—agentic workflow automation. The most sophisticated organizations aren't waiting; they're deploying multi-agent architectures now.