36Kr Publishes Interview on the “Moats” and “Hidden Reefs” of Embodied Intelligence

36Kr published an interview on Sept. 16, 2026, about the “moats” and “hidden reefs” of embodied intelligence.

For businesses involving AI that operates in the real world, operational and safety requirements matter alongside model performance. What assumptions statements and arguments make can therefore inform decisions by industry participants. 36Kr’s article addresses those issues, but assessing them specifically requires consulting the original text.

36Kr’s article is an interview that raises the question of how to assess the outlook for embodied intelligence after the release of GPT-6 Astra. However, in minimum short-brief mode, the statements themselves should not be reconstructed; the focus is on externally confirming the reception after the announcement, rather than on individual claims. See the original article for details. Taken together with our related articles, Turing Claims in an X Post That “FSD Is Beginning to Come Into View,” jidounten-lab Reports and Tesla Preparing Robotaxis for 24-Hour Operations; Current Service Runs From 6 a.m. to 10 p.m., discussion of AI that operates in the real world appears to be shifting from model performance alone toward the prerequisites for operation. The fact that 36Kr covered this issue suggests that advances in generative AI are becoming connected to the practical implementation of embodied systems. But technical achievements and business expectations need to be distinguished. For the industry, AI systems involving physical machines require more than model performance: operations at deployment sites, data collection, exception handling and safety validation are all at issue. In robotics and autonomous driving in particular, there can be a gap between a demonstration of capability and sustained operations, and the work needed to bridge it can become a source of competitive advantage. The issue raised by the interview is where that gap can be closed, as the discussion shifts from research and development alone to business design. The next questions to examine are which specific processes 36Kr’s “moats” and “hidden reefs” refer to, and how the release of GPT-6 Astra may change expectations on the implementation side. For now, it is more important to check the original statements and their context than to restate the article’s argument.

Japan has expertise in industrial robotics and control of physical machines, but the design of real-world operational data and ongoing validation can readily become competitive factors. By covering this issue, 36Kr has also made it a point for Japanese companies to examine where operational differences arise, rather than focusing only on model performance.