Research acceleration: The view inside OpenAI
Simon Willison’s Weblog Simon Willison ● Covered by 2 sources
OpenAI is talking about RSI, short for recursive self-improvement. It’s a clue the company thinks agents are moving from tools to something much bigger.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI seems to have picked a theme for the day: RSI, or recursive self-improvement. Simon Willison notes that the company is using the acronym without bothering to spell it out, while a separate essay, An Alien Mind, from chief scientist Jakub Pachocki, is on the same track. The message is hard to miss. OpenAI is openly framing its work around systems that can help improve themselves.
The piece also points to something more concrete than the acronym soup: OpenAI’s own research team is using coding agents. That alone says a lot about where the company thinks the useful work is happening. These tools are no longer just demos or side projects. They’re in the workflow of the people building the models.
Willison says 2026 has been the year agentic engineering really took off at OpenAI, and he says the clearest sign is a chart showing a sharp rise in AI spend per researcher. The interesting bit is the timing. He says there was a noticeable jump in late July.
His best guess is simple enough: that’s when internal employees got access to the model later released as GPT-6 Astra. If that guess is right, the chart is less about a random budget bump and more about what happens when a new model starts circulating inside the company before the public gets it. OpenAI, in other words, may be dogfooding its own future a bit harder than usual.
My take — AI-written commentary, not fact-checked reporting
This is the old Silicon Valley trick in a newer, shinier suit: call it self-improvement, hand everyone agents, and watch the spend line get a halo. The real tell is not the acronym, it’s that the company’s researchers are already working with the thing they’re building. That’s either discipline or the beginning of a very expensive feedback loop.
Read more about this at: Simon Willison’s Weblog
