Axios says Codex usage points to real AI-agent adoption

Axios reported that a new study from OpenAI with researchers at Columbia, Duke, and the University of Pennsylvania shows AI agents moving from hype into regular work. The report focuses on Codex usage and measures agent activity by tokens and task intensity. Axios said more than 80% of sampled users made at least one request equivalent to 30 minutes or more of human labor, while non-developers were the fastest-growing segment. The finding suggests agent adoption is spreading beyond software teams, even though many people still mostly interact with AI through basic chat interfaces.

Key details: Published June 25, 2026, The study covers OpenAI Codex usage, Researchers from Columbia, Duke, and the University of Pennsylvania participated, Axios reported non-developers were the fastest-growing user segment.

Why it matters: Agent adoption is the main question for the next phase of AI products; this gives a usage-based signal rather than another launch announcement.