OpenAI frames internal push as 'RSI day', touts research-team coding-agent acceleration
2026-09-06
Simon Willison reports OpenAI published a set of essays it internally branded around 'RSI' (recursive self-improvement), including 'An Alien Mind' by Chief Scientist Jakub Pachocki, alongside a piece describing how OpenAI's own research team has adopted coding agents through 2026. The piece includes a chart showing a sharp acceleration in AI spend per researcher starting in late July 2026; Willison speculates this coincides with internal employee access to the model later released as GPT-6 Astra.
Significance 3: Worth reading: a lab-strategy signal (OpenAI branding an internal 'RSI day' around AGI-adjacent essays) paired with a concrete internal usage-acceleration data point, but the only source is a curator (Simon Willison) relaying OpenAI's own essays rather than the primary text itself, so held at 3 rather than higher pending corroboration.
Implications · machine-drafted, not owner judgment
OpenAI is pairing an AGI-adjacent narrative ('RSI day', 'An Alien Mind') with a real internal metric — accelerating spend per researcher tied to agentic coding tools — which gives the framing more grounding than pure marketing, though the acceleration may simply reflect early access to a soon-to-ship model rather than a durable capability jump. For the owner's book timeline, this is another data point on whether OpenAI's AGI-by-December-2026 framing is substantive or promotional; worth a brief note in publication-window tracking without treating it as resolved either way.
- An independent account (ex-employee, journalist, competing lab) corroborating or disputing the claimed spend-per-researcher acceleration
- Whether OpenAI repeats or narrows the 'RSI'/AGI framing as December 2026 approaches
Sources
- curator Research acceleration: The view inside OpenAI retrieved 2026-09-07