# OpenAI Says Coding Agents Now Supply 3.1 Agent-Workdays of Effort for Every Human Workday

The company says it met its internal "automated research intern" target on schedule, with the median researcher consuming more than $600 a day of inference at list prices for application programming interface (API) access by mid-August.

- Published: 2026-09-07T06:22:00.097Z
- Canonical: https://polylog.news/ai/2026-09-07/openai-says-coding-agents-now-supply-3-1-agent-workdays-of-e
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [OpenAI](https://openai.com/index/research-acceleration-view-inside-openai), [Polylog editors](https://polylog.news)

OpenAI published internal telemetry on how its own researchers use coding agents and said it has reached the "automated research intern" goal it set last autumn. In [its post](https://openai.com/index/research-acceleration-view-inside-openai/), the company defines that milestone narrowly: a supervised system that takes a bounded objective, works across code and experiments, and returns results for a human to evaluate, including tasks that would take a skilled researcher a few days.

The headline number is a ratio. OpenAI says its research organization now draws 3.1 agent-workdays of effort for every workday of human labor. The supporting detail matters more than the ratio itself. At the start of 2026, the median OpenAI researcher used coding agents only lightly. By mid-August, according to the company, the median researcher was running agents daily and consuming more than $600 per day of inference at API prices, often in concurrent sessions. OpenAI also reports a secondary effect that is harder to manufacture: internal teams that held office hours to help researchers debug experiment infrastructure saw attendance decline through 2026, and at least one team stopped holding them. The [AI Post channel](https://t.me/aipost/8071) circulated the claim early in Asian and European hours, and [independent write-ups](https://www.unite.ai/openai-hits-goal-of-building-an-automated-research-intern/) reproduced the same figures from the company's post.

Every number here is self-reported and self-defined. OpenAI defined the target, chose the metric, and assessed its own progress against both, and "agent-workdays" is a measure with no external standard behind it. Dollar-denominated agent usage reflects spending, not productivity, and OpenAI pays its internal cost rather than the public list price. What is verifiable from the outside is the direction: the company is now willing to state a target date, [March 2028](https://www.itechpost.com/articles/237237/20260906/openai-reaches-its-automated-research-intern-goal-aims-have-automated-ai-researcher-march-2028.htm), for an automated artificial intelligence (AI) researcher, which converts a vague ambition into a falsifiable commitment.

## What this means

If agent-assisted research genuinely compresses experiment cycles, the binding constraint at a frontier lab shifts from researcher headcount to inference capacity, because each researcher's throughput scales with tokens rather than hours. That favors labs with captive compute (OpenAI, Google, Anthropic through its cloud partners) and disadvantages well-funded startups that buy capacity on the open market at retail rates. It also raises the internal cost floor for doing research, since a lab that does not spend heavily on agent inference is competing against one that does.

## What to watch

- Whether any lab outside OpenAI publishes comparable internal agent-usage data, which would turn this single vendor's self-reported metric into one with an independent baseline.
- Whether OpenAI's inference spending on internal research appears as a line item in its cost disclosures, which would let outsiders check the ratio against actual dollars.
- Whether the March 2028 automated-researcher target holds up through the next model cycle or quietly shifts, which is the clearest test of how much of this is measurement and how much is positioning.