Writing in Interconnects, the author pushes back on recursive self-improvement hype, noting that OpenAI and Anthropic are already deploying thousands of concurrent agents across their own operations. That internal scale is the concrete fact underneath the abstract debate.

The piece argues the San Francisco AI competitive culture creates an environment that amplifies every risk signal. Employees across both labs were already anxious about progress pace a year ago. Agents gaining product-market fit in early 2026 accelerated that anxiety. The author points out that forecasted risk timelines from 2023 and 2024 did not arrive on schedule, a pattern worth weighing against current claims.

The operating consequence is simple: when the loudest RSI warnings come from organizations running the most agents, the signal and the source are entangled. Watch whether this internal anxiety shapes policy positions, safety messaging, or open-source access debates at both labs.