In a post dated September 6, 2026, OpenAI said it has reached a goal announced last fall of having an 'automated research intern' by September of this year, which it defines as a system that can carry out well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days. The company said it is making progress toward an 'automated AI researcher' by March 2028. According to the post, OpenAI researchers are using coding agents throughout the day in concurrent sessions, with usage rising faster than among other internal teams, and are contributing code and running experiments more quickly. OpenAI stated that people still set research priorities and decide whether to scale, pause, or deploy systems. The company described pausing reinforcement learning training on its latest models after a 'Hugging Face incident,' hardening and red-teaming its research environments before resuming some workloads under stronger controls. OpenAI framed the disclosure as an effort to encourage public transparency about progress toward recursive self-improvement (RSI). Independent developer Simon Willison noted the piece coincided with a separate essay by Chief Scientist Jakub Pachocki, and speculated that a late-July spike in AI spend per researcher may reflect internal access to a model later released as GPT-6 Astra.
- OpenAI says it reached its 'automated research intern' goal by September 2026, targeting an 'automated AI researcher' by March 2028
- Researchers use coding agents in concurrent sessions daily, with usage outpacing other internal teams
- OpenAI paused RL training on latest models after a 'Hugging Face incident,' resuming some workloads under stronger controls
- The post is framed as voluntary transparency about progress toward recursive self-improvement (RSI)
What it means for you
OpenAI is telling us how its own staff use AI coding agents, and claiming those agents can now handle multi-day research tasks under human supervision. This is mostly a status report on how far frontier AI has come and how fast it's moving. It's interesting context, but it doesn't change anything you'd do this week to run your business.
Who should care
People following the pace and governance of frontier AI development, and software teams curious whether agentic coding tools are ready for heavier internal use.
Skip this if
You're a small business or non-technical reader looking for tools you can use today — this is a lab describing its own internal workflow, not a product you can adopt.
Sources: OpenAI, Simon Willison — read the original