Research acceleration: The view inside OpenAI
- ID
- 22040
- Status
- summarized
- Published
- 07 Sep 2026, 7:57 AM
- Fetched
- 07 Sep 2026, 9:06 AM
- Provider
- Simon Willison
- Category
- developer-ai
- Original URL
- https://simonwillison.net/2026/Sep/6/research-acceleration-the-view-inside-openai/
- Source URL
- https://simonwillison.net/atom/everything/
Summary
- Score
- 5.5
- Created
- 07 Sep 2026, 9:06 AM
- Tags
- Audience
- developersai_agent_usersai_ml_learners
What happened
Simon Willison links to an OpenAI piece on how their research team uses coding agents internally, noting that 2026 became the year agentic engineering took off at OpenAI. He highlights a chart showing a sharp spike in AI spend per researcher in late July, which he speculates coincided with internal access to the model later released as GPT-6 Astra. The post also references 'RSI day' (Recursive Self-Improvement) and a companion essay by Chief Scientist Jakub Pachocki called 'An Alien Mind.'
Why it matters
The concrete signal here is the late-July spend-per-researcher inflection point — if you're budgeting for agent usage in your own team, expect a step-change when a materially better model becomes available, not a gradual ramp. The post is thin on actionable detail beyond that chart, so treat it as a pointer rather than a playbook.
Discussion angle
What does the late-July spend spike imply for agent cost planning — should teams model a discontinuous budget jump when a new model generation lands, rather than linear growth?