OpenAI researchers find non-developer Codex usage grew 137-fold since August 2025 as agentic AI spreads beyond software teams

OpenAI researchers published a data-driven paper on June 25, 2026 analyzing usage patterns across Codex, their AI coding and agentic task product. The findings show a dramatic expansion beyond the tool's original software developer audience, with legal, finance, and recruiting teams now using Codex more than developers on many metrics.

What's new

The paper — "The Shift to Agentic AI: Evidence from Codex" by Drew Johnston, David Holtz, Alex Martin Richmond, Christopher Ong, Prasanna Tambe, and Aaron Chatterji — examines external personal accounts, organizational accounts, and OpenAI's own internal users. Key findings:

Growth:

• Active users grew more than fivefold in the first half of 2026
• Non-developer individual usage grew 137x since August 2025
• Organizational non-developer usage grew 189x over the same period
• Internal OpenAI non-developer adoption grew 12x

Adoption inside OpenAI:

• 98% of OpenAI employees now use Codex, up from roughly 40% in August 2025
• Codex usage has largely replaced business usage of ChatGPT among internal staff
• The median OpenAI researcher generated more than 50 times as many monthly output tokens in June 2026 compared to November 2025
• The median legal employee at OpenAI generated 13 times more monthly output tokens

Task complexity:

• More than 10% of users manage three or more concurrent Codex agents in a given week
• 26.6% of users use "skills" (structured multi-step agent capabilities)
• Since the start of 2026, the share of users submitting requests estimated to require more than eight hours of human expert work has increased nearly tenfold

Cross-industry adoption timing:

• Legal, finance, and recruiting crossed to majority use around April 2026 — more slowly than developers initially, but converging faster once adoption started

Context

Codex launched as a code-focused AI tool but has evolved into a general agentic platform capable of multi-step task management across professional roles. The paper covers the period from Codex's initial developer-focused release through the first half of 2026, capturing the transition from narrow coding tool to broad organizational workflow platform.

The 137x growth in non-developer individual usage is the most striking data point: it reflects a shift in who is using agentic AI rather than just how many. Previous AI productivity research tended to focus on developer output or knowledge worker writing assistance; this paper offers large-scale empirical evidence that agentic AI is reaching legal, financial, and operational functions at significant scale.

Why it matters

This is one of the first large-scale, data-backed analyses of agentic AI adoption patterns across an entire organization — including a major AI lab using its own tools. The 50x output token increase for researchers and 13x for legal staff are concrete productivity signals, though the paper's authors note the data is self-reported usage and output volume is a proxy, not a direct measure of work quality or business impact.

For enterprise buyers evaluating AI tool deployments, the cross-functional adoption data — legal and finance reaching majority use by April 2026 — suggests the transition from "developer tool" to "company-wide platform" can happen faster than anticipated once organizational barriers drop. For AI product teams, the 10x increase in requests for 8-hour+ tasks signals users are pushing agents into significantly longer-horizon work than a year ago.