OpenAI report: frontier enterprises now generate 8.3x more AI output per user than typical firms

OpenAI published two studies this week documenting a widening gap between enterprises that have embraced agentic AI and those still using it as a lightweight assistant, alongside evidence that Codex has overtaken ChatGPT as the primary source of enterprise AI output.

What's new

In a post titled "From assistance to execution: How enterprises put AI to work," OpenAI writes that "Frontier firms—those in the top 10% of AI usage each month—now generate 8.3× as many output tokens per active user as typical firms." That gap has roughly tripled since January, when it stood at 2.6x, according to the post.

The studies — an internal "Enterprise Signals" analysis of OpenAI's enterprise customer base and a companion working paper, "How Organizations Use AI: Evidence from ChatGPT" — surface several concrete findings:

• As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers, which OpenAI says points to a shift toward more substantive, delegated work rather than simple Q&A.
• Each week, 21% of active users at frontier firms use Plugins, versus 9% at typical firms; internally at OpenAI, 95% of employees use Plugins weekly.
• Since February, weekly active enterprise Codex users grew 108x in legal, 41x in sales, 41x in recruiting, and 26x in marketing — compared with 5x in engineering, the function most associated with AI coding tools.
• Usage is highest among early-career employees and declines with seniority.

Context

The report frames the shift as enterprises moving "from assistance to execution" — from asking AI questions to delegating full tasks to agents that use tools, create files, and produce work for review, via products like ChatGPT Work and Codex. OpenAI credits the widening frontier-firm gap to broader adoption of capabilities that connect agents to company context, internal tools, and repeatable workflows, rather than just heavier chat usage.

The functional breakdown is notable on its own: legal, sales, recruiting, and marketing are now growing their Codex usage far faster than engineering, suggesting agentic coding tools are being repurposed well outside software development inside large organizations.

Why it matters

The numbers give OpenAI's enterprise sales motion a concrete story to tell: adopt agentic workflows aggressively or fall further behind competitors who do. An 8.3x output gap between the top decile of customers and the median is a large enough spread that it functions as a competitive-advantage argument, not just a usage statistic — and it arrives as OpenAI pushes Codex and ChatGPT Work deeper into non-engineering functions.

It also offers an early data point on how agentic AI diffuses inside real organizations: unevenly, concentrated in a small set of "frontier" adopters, and increasingly outside the functions — like engineering — that were the first to get AI coding tools. If that pattern holds, the next phase of enterprise AI competition may be decided less by model capability and more by which companies build the internal permissioning, governance, and workflow habits needed to let agents actually execute.