A new OpenAI working paper on ChatGPT Enterprise is finally giving us real data instead of the endless parade of demo videos and cherry-picked success stories that have dominated the enterprise AI conversation for years. The research, authored by Aaron Chatterji, David Holtz, Neel Rakholia, Prasanna Tambe, and Gawesha Weeratunga, links actual ChatGPT Enterprise account records to usage metrics and worker role information—hard data we can actually reason about.

Why This Research Matters for Dev Teams

The authors make a pointed critique of what they call "demo theology" in enterprise AI adoption. We've all seen the polished case studies: Fortune 500s claiming 40% productivity gains, developers shipping twice as fast, support teams handling triple the volume. But these narratives often obscure the reality on the ground for the rest of us building and deploying this stuff. The working paper shifts focus toward observed behavior rather than self-reported outcomes, which is exactly the kind of empirical grounding infrastructure folks need when making build-versus-buy decisions.

What Organizations Are Actually Seeing

The research links account records directly to usage patterns across different worker roles, giving us a clearer picture of adoption curves and actual utilization rates. This matters because enterprise software purchases often don't match deployment realities—think about how many companies paid for expensive CI platforms only to have half their engineers continue using local builds. If ChatGPT Enterprise follows similar patterns, the business case some teams are building might be fundamentally flawed. The findings suggest that organizations with existing infrastructure for integrating AI tools into workflows see better returns. Firms already equipped with proper API management, authentication systems, and data pipelines can plug these tools in more effectively—rewarding those who've done the foundational work rather than those making first-time investments.

The Infrastructure Advantage

This is where it gets practical for the builders in the audience. If you're maintaining internal developer platforms, API gateways, or workflow automation systems, this research validates that investment. Companies with mature deployment pipelines and clear integration patterns appear better positioned to extract value from AI assistants than those scrambling to bolt everything together reactively.

Key Takeaways

  • Enterprise AI adoption success correlates strongly with existing organizational infrastructure maturity
  • Real usage data reveals gaps between demo claims and actual productivity gains
  • Worker role analysis shows which teams benefit most—and the patterns may surprise you
  • The "build it and they will come" approach to enterprise AI tools often fails without supporting systems

The Bottom Line

This research confirms what many of us have suspected: enterprise AI isn't a magic wand for organizations starting from scratch. If your company hasn't invested in the plumbing—authentication, API management, workflow integration—you're probably not going to see the same results as firms that built these foundations first. Start with infrastructure before you buy the flashy tools.