Gemini API Agents Get a Major Upgrade: Background Execution and MCP Support
On July 8, Google DeepMind added background execution and MCP (Model Context Protocol) support to the Gemini API's managed agents. Agents can now continue long-running tasks without requiring the user to wait, and integration with external tools and data sources has been standardized.
MCP is an open protocol proposed by Anthropic that provides a unified way for AI models to communicate with various tools and services. With the Gemini API now supporting it, agent interoperability across multiple AI platforms is expected to advance.
The addition of background execution has a particularly large practical impact. Until now, agentic apps required users to keep a session alive during long-running processes, but the ability to run tasks asynchronously significantly lowers the bar for real-world deployment.
Mistral AI Releases Robotics Navigation Model
French AI startup Mistral AI has released Robostral Navigate, a navigation model for robotics. Having signed contracts with major European industrial customers, the company is signaling a full-fledged entry into the Physical AI space.
Robostral Navigate is designed to navigate from a single camera input and is also drawing attention in the local LLM community. Mistral has so far focused its development on language models, but its move to expand into AI applications in the physical world is worth watching as an indicator of the European AI industry's strategic direction.
Crypto VC Paradigm Raises $1.2 Billion for AI
Paradigm, a venture firm founded to specialize in crypto investment, has formed a new $1.2 billion fund for non-crypto sectors, including AI and robotics. As the crypto sector shrinks, the movement to broaden investment targets into AI is accelerating.
The fund stands as a symbol of the massive capital shift from the crypto market to the AI market. That said, the investment frameworks of the two fields differ significantly, so how far crypto investing experience translates to AI investing remains a separate challenge.
Lateral Chip Stacking Breaks Through the AI Memory Wall
According to a technology trend reported by IEEE Spectrum, a new approach that stacks chips laterally (horizontally) is beginning to offer a fresh solution to the AI memory bandwidth problem. Conventional vertical stacking (3D stacking) faced significant constraints in thermal design and manufacturing processes, but leveraging horizontal connections reportedly enables more flexible memory scaling.
Since LLM inference performance is directly tied to memory bandwidth, this kind of hardware innovation has a direct impact on AI model efficiency. The timeline for mass production is unclear at this point, but it is a promising technological direction worth watching.
US Data Center Construction Faces a Severe Labor Shortage
According to Bloomberg, the US data center construction industry has enjoyed record backlogs over the past several quarters, but a shortage of skilled labor is becoming a bottleneck for growth. The lack of construction tradespeople has already caused delays in some projects, and industry executives hold a mix of optimism and concern.
The situation has highlighted a structural challenge: physical construction capacity cannot keep up with the explosive growth in AI infrastructure demand. Since workforce development takes time, the constraint is expected to persist in the short term.
Meta Enables AI Image Generation from Public Instagram Photos, Sparking Backlash
It has emerged that Meta has introduced a feature that lets users generate AI images based on public Instagram profile photos. As reported by the BBC and others, privacy concerns over the feature are spreading rapidly.
Because there is a possibility that your own face photos could be used in acquaintances' AI image generation, the feature is fueling debate over secondary use without consent. Meta has long faced criticism over where it draws the line between AI features and privacy, and this feature is also generating significant debate on social media.
---