OpenAI Reports Elevenfold Rise in Agent Research Use
OpenAI said on November 18, 2026, that agentic AI usage within its research organization has continued to increase exponentially since its first report on the topic, according to the company's post Research acceleration: The jagged frontier (OpenAI). The company stated: "Since our first report, agentic AI usage within the research organization has continued to increase exponentially."
Why This Matters
For people who build, buy or deploy agents, the figures offer a rare internal data point on how a frontier AI lab actually uses its own agentic tools in day-to-day research work, rather than a marketing claim about future capability. OpenAI reported that agent workdays per researcher rose from 0.5 to 5.7, a change the company itself describes as an elevenfold increase (OpenAI). That kind of jump, if representative of how agentic tools scale inside technical organizations, matters to enterprise buyers assessing how much agent capacity their own teams might need to provision, and to tool vendors building for concurrent multi-agent workflows.
Concurrency And Distribution
The report found that by mid-October, more than two-thirds of included researchers averaged a daily peak of at least four concurrent agents, while more than 10% averaged at least 16 concurrent agents (Riverfront AI). OpenAI also said the largest increases in agent usage occurred among teams focused on pretraining and posttraining, suggesting the heaviest adoption is concentrated in the stages of model development closest to core capability work (OpenAI).
What's Not Yet Confirmed
The publicly available excerpt of the report does not identify the authors by name beyond attributing the work to OpenAI, nor does it specify which models were tested, what benchmark methodology — if any — was used, or the sample size of researchers included in the measurements. OpenAI's own framing indicates the figures come from internal measurements of agent usage rather than a controlled benchmark experiment, and the available material does not state the report's methodological limitations.
What Comes Next
The report marks at least a second installment in OpenAI's internal tracking of agentic tool adoption among its own researchers, following an earlier version referenced in the November 18 post. Whether the company will publish more granular data — including team-level breakdowns, task types, or error rates for agent-driven research — remains unconfirmed based on the material currently available.


