OpenAI Demonstrates "The Shift to Agentic AI" with Codex Data

On June 26, OpenAI published a report titled "The Shift to Agentic AI" based on operational data from Codex. Analyzing actual usage patterns of the company's coding agent Codex, the report shows that users are clearly shifting from one-off questions toward agentic usage in which tasks are completed autonomously.

What makes the report notable is that it uses quantitative data from OpenAI's own product to back the claim that agentic AI is moving beyond the conceptual stage into practical use. It suggests that agent adoption in coding is beginning to reshape the structure of development workflows themselves.

Bruce Schneier on "AI and Liability"

Bruce Schneier, the renowned security expert, has published his thoughts on how liability should be handled for AI systems.

Schneier argues that as AI becomes more deeply embedded in decision-making processes, the areas that traditional legal liability frameworks cannot adequately address keep expanding. The question of who should bear responsibility for damage caused by AI malfunctions or harmful outputs—developers, providers, or users—has become urgent not only for regulators but also for corporate risk management.

The fact that Schneier, a long-influential voice in the security field, is tackling this topic in earnest is expected to add new weight to the debate over AI governance.

Analysis: "AI Inference Is Obviously Profitable"

Software engineer Sean Goedecke has published an analysis titled "AI Inference is Obviously Profitable."

Starting from the premise that AI inference costs are often overstated, the article examines the actual unit economics and argues that the inference business is structured to generate solid profits even today. Amid persistent market concerns about the profitability of AI companies, it claims—through a detailed analysis of cost structures—that the risk of losses is not as great as it appears.

That said, this view has not gained general consensus; profitability varies considerably depending on scale and use case, so it should be assessed with the caveat of "as things stand today."

AI Detection Tools Judge Paul Graham's 2013 Essay as "More AI-Like Than AI"

An intriguing case that calls into question the reliability of AI-generated text detectors has been reported. According to Nate of Ninjas & Robots, multiple local AI detection tools judged an essay Paul Graham wrote in 2013 to be "more AI-like" than actual AI-generated content.

This shows that AI detection tools tend to associate certain writing styles—clear, logical structure and polished vocabulary, for example—with AI generation. When excellent human writing is mistaken for AI output, it exposes a fundamental limitation of detection technology.

As the use of AI detection tools spreads in education and hiring processes, false positives like these can have real-world consequences. What is needed is both improved detector accuracy and operational designs that do not depend on them.

Disney Reimagines the Park Experience with AI Robotics

According to Bloomberg, The Walt Disney Company is accelerating R&D for next-generation theme park experiences. The company's R&D lab has developed a wooden manta ray prototype the size of a car, and the concept reportedly envisions a mechanism that would ultimately transform it into Gramma Tala, a character from the film "Moana."

In addition, the company is working on entertainment experiences that fuse AI and robotics, such as food-truck-style droids reflecting the Star Wars universe. The combination of Disney's IP (intellectual property) and cutting-edge technology could prove to be the key to differentiation in the theme park industry.

---