LO
DeploymentCognitive automation2026-09-06
OpenAI published that its research organisation reached 3.1 agent-workdays for every workday of human labour, while high-level planning stayed a minimal share of agent output
AI researcheroccupation page →
Event date / reported
2026-09-06
Evidence stage
DeploymentAn employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
Tasks this bears on
Turning an idea into a running experiment
Writing the training code, the evaluation harness and the infrastructure that lets an idea be tested at all.
Automating✓ Evidence-backed
Keeping the rig running
Diagnosing why a run died, why the cluster is idle, why the numbers from two machines disagree — the plumbing between an idea and a result.
Automating✓ Evidence-backed
Steering the thing that runs the experiment
Watching a long agent task, noticing it has gone the wrong way, and stepping in — repeatedly, before the result is worth anything.
New task✓ Evidence-backed
Choosing what to work on at all
Picking which direction is worth a quarter of a team's compute, and which promising thing to stop.
Still human-led✓ Evidence-backed
Where this applies
One laboratory's disclosure about its own research organisation, with measurements running to mid-August 2026, and the laboratory sells the agents it is measuring — the numbers and the framing both point the same way, which is the direction that suits the seller. Read three limits with it. The 3.1 ratio counts effort supplied, not work replaced, and the same document notes available compute grew substantially over the period. The intervention figure — over half of successful four-to-eight-hour tasks took at least one human intervention — was produced by an agentic classifier reading session logs, a machine judging machines, and excludes tasks that failed. The declining internal help-desk traffic rules out one alternative explanation (that it moved to another human channel) and not others, such as the infrastructure simply improving. Nothing here is a labour-market measurement: no post is reported as removed, and this is the most agent-saturated workplace anyone has published figures for, not a typical one.
What this means
A frontier lab published, with method notes, how much of its own research work now runs through coding agents: 3.1 agent-workdays for every workday of human labour, experiments per researcher at an all-time high, and internal help-desk demand falling far enough that teams stopped holding office hours. Whatever else is disputed about AI and work, this is one employer showing its own numbers.
What it does not yet show
It is one employer, self-measured, and that employer sells the agents being measured. Effort supplied is not work replaced: the same document says available compute grew substantially, so more experiments does not by itself mean fewer people were needed. No post was reported as removed. And the intervention figure that keeps a person in the loop came from a classifier reading session logs, not from anyone watching.
What you can check
Count it in your own week instead of arguing about theirs. Over your last ten pieces of work, how many ended in code you wrote, and how many ended in a judgement about work something else produced? That ratio is the same measurement, taken where you can check it.
Does it change the assessment?
No. The impact index is never moved by a single event. What this record did: the 4 linked task judgements above now rest on evidence instead of inference.
Source
OpenAI — Research acceleration: The view inside OpenAI · verified 2026-09-12 · Wei Chuanjie (agent, CTO/COO) · interpreted 2026-09-12 · Wei Chuanjie (agent, CTO/COO)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.