Legal teams writing code. Recruiters building engineering workflows. Finance departments tackling tasks that used to require a CS degree. OpenAI’s latest research, released June 25, shows that its agentic AI tool Codex isn’t just making workers faster. It’s making them do entirely different jobs.

The numbers behind the blur

OpenAI’s Economic Research team tracked Codex usage between August 2025 and June 2026. Non-developer token consumption surged 137-fold among individual users and 189-fold among groups. By June 2026, Codex accounted for 99.8% of all weekly output tokens.

Over 25% of Codex work performed by business-function employees involved engineering or coding tasks. People in Legal, Finance, and Recruiting were effectively crossing into technical territory that would have required hiring a developer just a year earlier.

About 80.6% of surveyed users made at least one Codex request estimated at over 30 minutes of equivalent human work. Roughly 25.6% of users exceeded eight hours in delegated task time.

OpenAI’s April 2026 “AI Jobs Transition Framework” categorized occupations by vulnerability to automation and found that AI can theoretically handle around 90% of tasks in the highest-risk roles. But actual usage was considerably lower, with workers engaging AI tools for less than a quarter of that theoretical potential.

A June 2026 EU framework extension noted that Europe has lower shares of high-automation-risk employment compared to the US. That geographic divergence matters for crypto companies deciding where to locate talent and operations.

When 99.8% of output tokens flow through a single vendor’s agent, that’s a concentration risk. Projects building on AI-augmented workflows need to consider what happens when the API goes down, the pricing model changes, or the model’s outputs degrade.