OpenAI Researchers Now Use 3.1 Agent-Days Per Workday — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI announced it hit its 'automated research intern' goal, with researchers now averaging 3.1 agent-workdays per human workday — the post says usage has rapidly outpaced other OpenAI teams since the start of 2026
- Top OpenAI researchers spend $7,000+/day on tokens, and per Robert Yang's estimate researchers average roughly $200,000/year on coding agents at API pricing (less internally)
- The median OpenAI researcher spent $0/day on coding agents as recently as February 2026, per Charlie O'Neill, underscoring how recently this adoption curve went vertical
- OpenAI said it is on track for a fully autonomous AI researcher by March 2028, with biggest agent use cases being research and infrastructure code, technical review, and tracking training/RL/eval runs — while high-level research design still requires humans, per Chinmay
- OpenAI reported 7x code changes per contributor and 1.6x experiments per experimenter over the same period, with Dylan Patel noting OpenAI employees now match SemiAnalysis employees in AI use after previously trailing them
- Simon Willison flagged an unexplained mid-July 2026 spike in OpenAI token spend, speculating whether GPT-6 Astra becoming available to employees triggered it
Why it matters: OpenAI's per-researcher coding-agent spend ($200k/year at API prices, with top users at $7k+/day) reframes frontier AI research as compute-bound rather than labor-bound, with the March 2028 fully-autonomous-researcher target formalizing that trajectory — a dynamic that directly raises infrastructure demand on GPU and cloud providers serving OpenAI.
Ask SkimNews
