OpenAI published internal research showing enterprise AI adoption has shifted from question-answering to autonomous task execution. Frontier adopters, defined as the top quartile of enterprise users, run multi-step agentic workflows using ChatGPT and Codex, delegating code review, data analysis, and document drafting to models that operate with minimal human checkpoints. The gap between frontier firms and average adopters is widening, not narrowing.
The report is worth reading for its taxonomy of adoption stages, not just its conclusions. OpenAI breaks enterprise usage into distinct phases: assistance, augmentation, and execution. Most companies are still in augmentation. The firms in execution mode are using Codex to automate software development pipelines and deploying ChatGPT agents that chain tools across APIs without per-step human approval. Those structural details explain why productivity gains are uneven across industries.
The next pressure point is governance. As enterprises move into execution workflows, the question shifts from capability to accountability: who owns the output when an agent acts autonomously across systems. OpenAI acknowledges this without resolving it, which makes the full report a useful map of where enterprise AI is heading and what organizational infrastructure most companies have not yet built.
[READ ORIGINAL →]