Gemini Co-Lead Noam Shazeer Moves to OpenAI
It has been revealed that Noam Shazeer, co-author of the "Attention Is All You Need" paper and co-lead of Google's Gemini models, is joining OpenAI. Shazeer returned to Google from Character.AI in 2024, in a deal reportedly worth around $2.7 billion. This marks another move just two years after his return to Google.
It is the second major AI talent shift this year, following Andrej Karpathy's move to Anthropic. One of the original creators of the Transformer architecture joining OpenAI is drawing attention as a move that will strengthen the company's research and development capabilities.
Microsoft Expands OpenAI Model Sales in China, Building Presence in the AI Market
According to Bloomberg, Microsoft is expanding its AI business in the Chinese market through sales of OpenAI models. As technology restrictions between the US and China grow increasingly strict, Microsoft is leveraging its unique position as a provider of OpenAI models to Chinese companies.
The company's AI business in China appears to be growing in a way that sidesteps the regulatory barriers faced by other US tech companies. However, detailed revenue figures and clients have not been reported, and the business could be affected depending on how the geopolitical environment shifts going forward.
Reports of "Claude" Outages Mounting — Users in Japan Affected
According to an ITmedia AI+ article, users of Anthropic's AI assistant "Claude" are increasingly reporting issues such as "it's not working" and "is it down?" Users in Japan have reported the service becoming unavailable, though details on the specific recovery time and cause remain limited at this point.
Since Claude is widely used in the developer community, the impact of the outage is significant. Users are advised to check the official status page and announcements from Anthropic.
Databricks Releases Agent Management OSS "Omnigent"
In a keynote at Data + AI Summit 2026 held in San Francisco, Databricks co-founder and CTO Matei Zaharia introduced "Omnigent," an open-source project released just this past weekend. Omnigent is a layer called a "meta-harness" that sits on top of individual AI agents, providing a framework for coordinating and managing multiple agents.
As gains in individual agent performance show signs of plateauing, how to design coordination between agents is considered the key to practical adoption. Omnigent takes a direct approach to this challenge, and its ecosystem development going forward will be worth watching.
Examining METR's "19% Slower with AI" Finding — Are We Even Measuring Development Speed Correctly?
A verification article examining METR's (Model Evaluation & Threat Research) finding that "experienced developers were actually 19% slower when using AI coding tools" has been published in Japan's technical community.
The article cross-checks the subjective feeling that "AI made development faster" against primary data, and concludes that the real issue is that "before debating whether things got faster or slower, developers themselves aren't even properly measuring their own development speed." This gap between perception and actual measurement points to a structural problem in assessing the effectiveness of AI adoption.
---