Mavericks AI News Passes 85,000 Subscribers; February 2 Issue Explains Genie 3

Mavericks, Inc., the developer of the AI video generation service NoLang and other advanced AI products, has published the latest issue of Mavericks AI News, a generative AI newsletter that has passed 85,000 subscribers and ranks among the largest of its kind in Japan¹. The issue was delivered on February 2.

This issue explains three topics: Genie 3, the world model that Google DeepMind released in the United States, and what its simulated-space generation means for the path to AGI; Agentic Vision, a Gemini 3 Flash capability that sharply improves the accuracy of image analysis; and MCP Apps, which lets users operate external applications directly inside Claude. The commentary comes from an engineering team that works on AI product development every day, and it focuses on what AI adoption actually means in practice.

Mavericks AI News: field knowledge from the team behind the AI video generation service NoLang

Mavericks, Inc. is a Japanese startup that offers the generative AI product NoLang. About a year and a half after launch, NoLang has more than 150,000 registered users and has been adopted by over 60 corporate customers, and the team runs its own product at the front line of AI-driven development.

The generative AI industry changes extremely fast. What can generative AI actually do today, and what makes the AI services drawing attention in the market valuable? Mavericks places importance on the substantive knowledge that can only be gained on the ground in product development, which is not visible from the outside.

For Mavericks AI News, a team working daily at the front line of AI development selects genuinely useful topics from a vast amount of information. Each issue adds original commentary so that even readers with no prior AI knowledge can picture concretely how to apply the news from the next day onward. It is delivered every Monday.

Highlights from this week’s issue (delivered February 2)

Feature 1: Google DeepMind releases the world model Genie 3 in the United States, from robotics and self-driving to the path to AGI

Google DeepMind has opened Project Genie, an experimental project that lets users try the previously announced world model Genie 3, to users in the United States. A “world model” is an AI that can generate a simulated space like the one depicted in the film The Matrix, and users can move freely through the generated 3D space with a PC keyboard.

From a text or image input alone, a world at 720p resolution and 24fps is generated in 10 to 30 seconds, and users can explore it while controlling characters, cars and various objects. The issue covers the detailed specifications and application scenarios: near-term uses in game development and education, a medium-term outlook as a training environment for robotics and self-driving, and the longer path to AGI.

Feature 2: Agentic Vision released, sharply improving Gemini’s image analysis with the power of agents

Google has released Agentic Vision, a new Gemini 3 Flash feature that improves the accuracy of image analysis. When the model analyzes an image, it runs appropriate code to arrive at a more accurate answer.

For example, when reading a part number printed on a chip in a photograph of a circuit board, the model crops the chip area and enlarges and rotates it, so it can identify small characters accurately in cases that used to be difficult. The issue explains how the Think-Act-Observe cycle works and shows examples such as counting fingers and building a complex chart. It also compares processing time with o3 and considers the feature’s potential as an effective approach to the problem that “AI is blind.”

Feature 3: MCP Apps, the first MCP extension, is released, making it possible to operate apps directly inside Claude

Anthropic has announced that the features of external applications such as Figma, Slack and Box can now be used directly in the Claude chat screen. Users can review a flowchart created in Figma on the spot, or have Claude compose a message, show it as a Slack draft, and then edit and send it right there.

The feature can be described as Claude’s counterpart to Apps in ChatGPT, which OpenAI announced last October. The issue covers the state of integration with more than 100 apps and the details of the MCP Apps standard, examines practical cases such as Google Drive and Box, and analyzes the outlook from the perspective of how long users stay in AI chat services.

Outlook

As a product-out generative AI startup, Mavericks, Inc. considers it an important mission to return to society the knowledge gained from developing advanced products in-house. Through Mavericks AI News, the company will continue to deliver reliable, practical information to every business professional, developer and DX lead working to open up the AI era.