OpenAI published “Research acceleration: The view inside OpenAI” on Sept 6, the first formal public measurement of how agentic systems are accelerating its own research. As of mid-August the research organisation uses 3.1 agent-workdays of effort for every standard 8-hour workday of human labour; the median researcher was already past $600 per day of agent-inference at API prices and the 90th percentile past $7,000 a day; experiments per active researcher hit an all-time high in August since tracking began in Jan 2025 (OpenAI, Sept 6).

The numbers connect to the goal OpenAI set last fall: fielding an “automated research intern” — a system that carries out well-defined research tasks, including ones a skilled researcher would need days for, under human direction. The company says measurements now show it has reached that milestone, and that it is making strong progress toward an automated AI researcher by March 2028, with a call for public RSI-progress tracking to become the norm.

The disclosure also quantifies what happened when safety controls bit: after the July 20 container-breach hardening, RL training paused for about two weeks, and after Aug 7 evidence that Astra-class compute may carry critical cyber capabilities, Astra-class GPU allocation fell 59.2% for a week while other model classes rose 17.2% — offsetting about 85% of the Astra-class decline. Success rates rose January-to-July, but more than half of successful 4-8-hour tasks still involved one or more human interventions in the last six months.

For Indian enterprises the usable number is the 3.1x agent-workload with its caveats: if a frontier lab delegates 3.1 days of agent effort per employee-day but still steers more than half of the longer tasks, Indian IT’s agent-led delivery is best modelled as augmentation with gates, and the $600-to-$7,000-per-day inference band gives CIOs a concrete budget reference for Q4 agent fleets.

Today is also the India deadline: MeitY’s National e-Governance Division reopened AI/ML vendor empanelment for public-sector work and proposals are due today, Sept 7, with technical bids opening Sept 8; Round 1 already selected TCS, NEC India and four others (Moneycontrol; Storyboard18). Markets rest at the Sept 4 close — Sensex 76,515.43, Nifty 23,897.70 (weekly -1.15%), INR 94.49, Brent ~$95 — before the week’s first session, which opens under new NSE pre-open order rules.