The measurement
OpenAI published a post titled "Research acceleration: The view inside OpenAI" on 6 September 2026. Its central claim, quoted at unite.ai: "According to our measurements, we have now reached the goal, announced last fall, of having an automated research intern by September of this year." The same post reports that the company's research organization now logs 3.1 agent-workdays of effort for every workday of human labor, drawn from internal usage during August 2026.
The intern label is defined narrowly. Per the company's framing, an automated research intern carries out well-defined research tasks under human direction, including tasks that would take a skilled human researcher several days. The 3.1-to-1 ratio counts agent effort against human effort inside OpenAI's own research organization; it is a measure of workdays logged, not of output quality or of tasks completed without supervision.
Both the metric and the milestone are self-reported. OpenAI set the September 2026 deadline itself, ran the measurement itself, and declared the goal met itself. Simon Willison, writing at simonwillison.net, notes that the post and a companion essay treat "RSI" — recursive self-improvement — as a framing, and that the post does not bother to expand the acronym, assuming readers already know it. That assumption is itself a signal about the intended audience.
The roadmap
The timeline predates this announcement. OpenAI flagged September 2026 as its internal target for a research intern and 2028 for a fully automated researcher when it laid out the roadmap publicly last year. The new post is a progress report against those self-imposed dates: the intern milestone is declared reached on schedule, and the company now states a specific month, March 2028, for the researcher.
The step from intern to researcher is a change in supervision, not scale. The intern handles well-defined tasks under human direction; the researcher, per the company's target, is fully autonomous. Eighteen months separate the two milestones. That suggests OpenAI treats removing human direction from research work as the harder problem, and is pacing the claim accordingly rather than collapsing the two categories into one announcement.
Chief Scientist Jakub Pachocki published a companion essay, "An Alien Mind", alongside the progress post. Willison's reading is that recursive self-improvement functions as the company's new AGI framing. The pairing matters: the intern milestone is an operational claim about tooling, while the essay frames the same trajectory in terms of systems improving the systems that produce them. The March 2028 date sits between those framings.
What the data shows
The post includes internal usage data on how OpenAI's research team uses coding agents. Willison highlights a chart showing a sharp acceleration in AI spend per researcher in late July 2026. His stated guess is that the jump marks the point when internal employees gained access to the model later released as GPT-6 Astra. He also describes 2026 as the year agentic engineering took off at the company.
What the data does not show is external verification. The 3.1 figure comes from OpenAI's own instrumentation of its own workforce; no independent audit, task-level breakdown, or quality control accompanies the number in the published material. The record is silent on how agent-workdays were weighted against human ones. Coverage at cryptobriefing.com reports the figures as published, without independent measurement of its own.
The open question is whether the ratio continues to climb toward the March 2028 target, and what a fully autonomous researcher would be measured against once human direction drops out of the denominator's definition. Two signals are worth tracking: whether the spend-per-researcher curve holds its late-July slope, and whether OpenAI publishes task-level data to substantiate the intern claim it has now declared met.
Sources
- simonwillison.net — 3.1 Agent-Workdays for Every Human One: OpenAI Measures Its Own Automation
- cryptobriefing.com — OpenAI reveals AI systems now perform 3x the research effort of its ... (2026-09-06)
- officechai.com — OpenAI Says It Has Reached Its Goal Of Having An ... (2026-09-06)
- unite.ai — OpenAI Hits Goal of Building an ‘Automated Research Intern’ (2026-09-06)
- New Horizon daily digest, 2026-09-07 (topic provenance)