Research publication Provisional 55% confidence first seen
OpenAI reported that its research team is using coding agents to change how experiments are planned and run, aiming to increase speed and capability on complex tasks. The coverage also links internal research acceleration to higher AI spend per researcher and wider adoption of agentic engineering practices across the team.
The daily briefing
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
6 September 2026
OpenAI’s internal research playbook is getting agentic—and the company is tying it to money, not magic. In two pieces reflecting how coding agents are reshaping day-to-day work, OpenAI describes research-focused agents running recursive loops to speed up experiment design and execution, with a late-July 2026 jump in AI spend per researcher framed as the lever behind faster “experiment velocity.” The same theme shows up elsewhere: Meta FAIR formalized that idea into product form with AI Research Preference Models (RPMs), using inference-only ranking to decide which unexecuted candidates deserve GPU time, cutting time-to-baseline on a 24-hour test from a day to about 15–15.5 hours while still posting new highs like WinoGrande at 94.1% and SVAMP at 95.7%.
Read the full briefing →