RSI and the rise of agentic engineering
OpenAI has published new insights into its internal development processes, focusing heavily on what it calls Recursive Self-Improvement (RSI). In a new release titled "Research acceleration: The view inside OpenAI," alongside an essay called "An Alien Mind" by Chief Scientist Jakub Pachocki, the company details how its own research team is utilizing coding agents to accelerate development.
While full technical details are limited so far, the primary takeaway is that 2026 has become the year of "agentic engineering" at OpenAI. Developer Simon Willison highlighted an internal chart from the release showing a significant spike in AI spend per researcher beginning in late July. Willison hypothesizes that this jump in internal compute usage coincides with employees gaining early access to the model that was subsequently released as GPT-6 Astra.
What agentic workflows mean for your operations
OpenAI's internal shift matters for small and mid-sized businesses because it validates a crucial market trend: the transition from using AI as a basic chat interface to deploying it as autonomous, task-oriented agents. If the company building the world's leading models relies heavily on coding agents to accelerate its own engineering, the practical value of agent-based workflows is undeniable.
For SMBs, "agentic engineering" translates directly to agentic operations. Instead of relying on human staff to manually prompt an AI for every step of a process, businesses are deploying specialized agents to handle discrete, repeatable workflows—such as parsing inbound vendor invoices, migrating customer data between systems, or executing daily compliance checks.
The spike in OpenAI's internal AI spend per researcher also highlights a reality of modern automation. As human employees begin offloading complex, multi-step tasks to AI agents, software and compute costs per employee will naturally rise. However, this is fundamentally an investment in operational leverage. Just as OpenAI researchers use agents to accelerate complex model development, your administrative teams can use specialized AI agents to dramatically increase the volume and accuracy of the workflows they manage every day.