OpenAI's new work research matters because it measures agent usefulness in task duration, parallel runtime, and departmental adoption instead of only capability demos.
What OpenAI Published
On June 25, 2026, OpenAI published How agents are transforming work, describing agentic AI as a shift from short interactions toward delegated, long-horizon tasks.
OpenAI reports that by May 2026, 80.6% of sampled individual users made at least one Codex request estimated to exceed 30 minutes of human work, 70.2% crossed one hour, and 25.6% crossed eight hours. It also says Codex became the primary AI tool for every department inside OpenAI and that non-developer adoption grew faster than developer adoption across user groups.
Why This Capability Signal Is Strong
Most agent discourse still focuses on whether a model can solve a task. OpenAI is focusing on whether users actually delegate work that stretches over meaningful time horizons and whether that work crosses occupational boundaries.
That is a stronger signal for zero-human companies. A model is economically important when it can absorb enough operational time that people restructure how work gets done around it.
Why Departmental Spread Matters
The most interesting statistic is not only that engineers use Codex heavily. It is that legal, finance, and recruiting became majority users as well, and that business-function work done with Codex often included technical execution.
In other words, agent capability is no longer confined to a coding surface. It is becoming a general mechanism for crossing task boundaries that previously required handoffs between departments.
The Take
OpenAI's latest research is a meaningful capability signal because it shows that agent work is getting longer, more parallel, and more widely distributed across job functions.
The more often companies can hand off multi-hour tasks to agents instead of one-turn questions, the closer zero-human companies get to replacing departments with coordinated agent labor.
Related: See our previous research on ChatGPT Work, OpenAI's Codex labor thesis, and GPT-5.6 Sol.