The Shift to Agentic AI: Evidence from Codex

Drew Johnston, David Holtz, Alex Martin Richmond, Christopher Ong, Prasanna Tambe, Aaron Chatterji · arXiv · 25 Jun 2026

Read original sourceOpen access

Where this sits in the debate

Will AI actually boost productivity and growth? · its finding supports the claim (why studies differ)
How fast will the impact arrive? · its finding supports the claim (why studies differ)

Direction of the finding as read from the abstract by an AI model (AI-written).

Key takeaway AI-written

Analyzing OpenAI Codex telemetry across personal, organizational, and internal-OpenAI users, the authors document active users growing more than fivefold in the first half of 2026 (fastest outside software developers), over 10% of users managing 3+ concurrent agents weekly, 26.6% using shareable skills, and a near-tenfold rise in the share of users submitting tasks estimated at 8+ human-hours. Inside OpenAI, Codex has largely displaced business ChatGPT use, offering an early look at what near-frontier agentic adoption implies for workflow restructuring — though external organizational uptake remains far more uneven, underscoring the diffusion gap central to AI-productivity forecasts.

The paper in three steps AI-written

Written by an AI agent from the paper's abstract, which is the primary source and is quoted below.

The question

How is agentic AI, which takes actions on a user's behalf, changing how people work, judged from usage of OpenAI's Codex tool?

The approach

The authors analyse Codex usage data through an automated, privacy-protecting pipeline, comparing external personal-account users, external organizational users and OpenAI's own staff, and track usage volume, sophistication, request complexity and output.

The findings

Active users grew more than fivefold in the first half of 2026. Within OpenAI, Codex use is nearly universal and has largely replaced business use of ChatGPT. Over 10% of users run three or more concurrent agents weekly; the share submitting tasks estimated at over eight hours of human work rose nearly tenfold. Median output tokens rose 13-fold for legal staff and 50-fold for researchers.

How to read it: observational. Analysis of real-world firm, worker, or market data — realistic settings; causal claims depend on the identification strategy.

Limitations and counterpoints AI-written

The data come from one company's product and its own employees, an early-adopting group, and output tokens measure usage rather than productivity or job outcomes.

Abstract

We analyze usage data from OpenAI's Codex tool to present large-scale evidence of how agentic AI technology, which can take actions on a user's behalf, changes how people work. We use an automated, privacy-protecting pipeline to contrast usage across three populations: external personal-account users, external organizational-account users, and workers within OpenAI. We find that agentic AI usage is growing rapidly: the number of active users has grown more than fivefold in the first half of 2026, with the most rapid increase occurring outside the initial audience of software developers. Uptake is uneven: within OpenAI, Codex usage is nearly universal and has largely replaced business usage of ChatGPT. We document a similar shift to agentic tooling outside OpenAI, particularly within organizations, although external adoption remains lower and more uneven. In addition to headline usage figures, we observe measures of sophistication, and find that a growing number of users have used Codex to change their workflows substantially. More than 10% of users manage three or more concurrent Codex agents at some point each week and that 26.6% use skills, which allow users to share instructions for complex workflows. Alongside these changes in usage practices, request complexity has increased: since the start of the year, the share of individual Codex users who submit at least one request for a task estimated to require more than eight hours for an experienced human to complete has increased nearly tenfold. Concurrently, output has grown rapidly -- in June 2026, the median OpenAI employee in a legal role generated 13 times more monthly output tokens across Codex and ChatGPT than they did in November 2025, while the median researcher generated more than 50 times as many. We conclude by discussing the implications of these patterns for productivity, job reorganization, and workforce restructuring.

About this publication

Preprint. A preprint — posted publicly before or without journal peer review; findings may still be revised.

Published: 2026 (4 months ago)
Citations: Not recorded for this source

Citation counts come from OpenAlex; none are recorded for this source — absence of a count is not absence of citations.