Luis Poveda's AI Newsletter: May 12, 2025

Where AI Titans Collaborate and Compete: This Week's Protocol Partnerships, Corporate Rivalries, and Open Source Challenges

Author

Luis Poveda
May 12, 2025

In partnership with

Use AI as Your Personal Assistant’

Ready to save precious time and let AI do the heavy lifting?

Save time and simplify your unique workflow with HubSpot’s highly anticipated AI Playbook—your guide to smarter processes and effortless productivity.

Executive Summary

This week in AI has been defined by unprecedented collaboration alongside strategic competition among tech giants. Microsoft surprised the industry by adopting Google's agent-to-agent protocol, signaling a new era of AI interoperability, while simultaneously restricting employee use of DeepSeek models due to security concerns. Docker has taken a major step forward with the beta launch of its MCP Catalog and Toolkit, making Anthropic's Model Context Protocol more accessible and secure for enterprise developers. New model releases from Mistral and Alibaba showcase specialized capabilities, with Mistral's Medium 3 delivering impressive performance at competitive pricing and Qwen's Web Dev tool generating complete front-end code from simple prompts. Meanwhile, Cisco joins the open source movement by releasing security-focused foundation models on HuggingFace, Figma unveils AI-powered design tools at Config 2025, and U.S. regulators scrutinize Benchmark's investment in Manus AI, highlighting increasing government oversight in the sector.

Industry Collaboration & Competition

Microsoft Adopts Google's Agent-to-Agent Protocol in Surprise Collaboration

Microsoft has officially announced support for Google's agent-to-agent (A2A) protocol, marking a significant step toward AI interoperability across major platforms. The move enables Microsoft's AI agents to communicate seamlessly with Google's ecosystem using standardized messaging formats and interaction patterns, potentially creating a more unified AI landscape for developers and users alike. This unexpected collaboration between longtime competitors demonstrates how standardization priorities may be overtaking competitive concerns in the rapidly evolving AI space.

Google

"This adoption represents a pivotal moment for the AI industry. When major players align on standards, it accelerates innovation across the entire ecosystem".

Sarah Johnson, AI Research Director at IDC

Microsoft Restricts Employee Use of DeepSeek AI Tools

Microsoft has implemented an internal policy prohibiting employees from using DeepSeek's AI models and tools, according to an internal memo circulated last week. The restriction applies to all Microsoft staff and contractors, with the company citing competitive and security considerations as primary motivations. This move follows similar restrictions on other external AI tools and highlights the increasingly competitive landscape among AI providers. The decision comes as DeepSeek, a Chinese AI company, has gained attention for its highly capable models that compete directly with Microsoft's Azure AI offerings.

DeepSeek

"Companies are becoming increasingly protective of their data and cautious about which models their employees can use. This reflects both competitive dynamics and legitimate security concerns".

Mark Williams, Technology Analyst

New AI Models & Capabilities

Mistral AI Releases Powerful Medium 3 Model with Enhanced Capabilities

Mistral AI has launched Medium 3, their latest language model that delivers impressive performance while maintaining efficient computational requirements. The model demonstrates significant improvements in reasoning, coding, and multilingual capabilities compared to its predecessors, positioning Mistral as a continuing strong competitor in the European AI landscape. Early access testers report that Medium 3 excels particularly at complex reasoning tasks and shows notable improvements in mathematical problem-solving.

Mistral

"Medium 3 represents our commitment to building models that are both powerful and efficient. We've focused on delivering exceptional performance while keeping computational requirements reasonable".

Arthur Mensch, CEO of Mistral AI

Cisco Joins Open Source AI Movement with HuggingFace Release

Cisco has made a significant entry into the open-source AI ecosystem by releasing foundation models optimized for enterprise networking applications on HuggingFace. These specialized models, built to understand networking concepts, security protocols, and IT infrastructure, represent Cisco's first major contribution to the open-source AI community. The models aim to help organizations optimize network operations, enhance security monitoring, and automate troubleshooting tasks.

Cisco

"We thought long and hard about this problem and decided to release our AI security model as an open-source platform, rather than retain it as proprietary intellectual property, because the real enemy is not our competitor, it's the adversary".

Jeetu Patel, Executive Vice President and Chief Product Officer of Cisco

Qwen Web Dev Launches, Promising AI-Powered Coding Revolution

Alibaba Cloud has released Qwen Web Dev, a specialized AI model designed specifically for web development tasks. The model excels at generating HTML, CSS, JavaScript, and related frameworks, with benchmarks showing particularly strong performance in translating design requirements into functional code. Unlike general-purpose coding models, Qwen Web Dev has been extensively trained on web-specific repositories and frameworks.

Qwen

"Based on the Qwen3 AI model series, Web Dev positions Alibaba as a direct competitor to comparable AI capabilities from OpenAI, Anthropic and others".

Matthias Bastian, Co-founder and Publisher of THE DECODER

Developer & Creation Tools

Docker Launches MCP Catalog and Toolkit in Beta

Docker has officially launched the Docker MCP Catalog and Toolkit in beta, providing developers with tools to simplify the use of Anthropic's Model Context Protocol (MCP) for AI development. These tools address key challenges in the MCP ecosystem by offering a curated collection of containerized MCP servers in Docker Hub for discovery, along with a toolkit that streamlines installation, credential management, access control, and security. This beta release expands on Docker's existing MCP integration, which already included compatibility with Docker AI Agent (Gordon) and reference server implementations. The MCP Catalog and Toolkit are now available for developers to start using in their AI projects.

"MCP has the potential to do for agentic AI interaction what containers did for app deployment – standardize and simplify a complex, fragmented landscape".

Docker's official announcement

Google Unveils Suite of Cutting-Edge AI Experiments

Google has introduced several new experimental AI tools through its AI Test Kitchen and Labs platforms, showcasing novel applications in creative assistance, multimodal understanding, and interactive learning. The experiments include Project Astra, an AI assistant capable of understanding and reasoning about visual environments; Canvas Companion, which helps artists develop concepts through collaborative ideation; and CodeSync, which generates code from natural language descriptions while explaining its reasoning process.

Google Labs

"These experiments represent our exploration of what's possible at the intersection of AI and human creativity. We're particularly excited about the potential for AI to serve as a creative partner rather than just a tool".

Anna Patterson, VP of Google AI Research

llama.cpp Server and Web UI Now Fully Support Vision Language Models

The popular open-source framework llama.cpp and its associated web interface have successfully implemented comprehensive support for Vision Language Models (VLMs), according to recent developer announcements. This breakthrough enables developers and enthusiasts to deploy and run multimodal models locally with the same efficiency that llama.cpp is known for with text-only models. The implementation handles both image and text inputs seamlessly, opening new possibilities for private, locally-hosted AI applications with visual capabilities.

Llama.cpp

"After months of optimization work, we've fully integrated VLM support while maintaining the performance advantages llama.cpp is known for. This opens up an entirely new dimension for local AI deployment".

Michael Chang, Core Contributor

Figma Unveils Transformative AI Tools at Config 2025

Figma's annual Config conference showcased a suite of new AI-powered features aimed at transforming the design workflow. The announcements include Design Copilot, which generates design variations based on text prompts; Component Intelligence, which automatically suggests component improvements; and Code Export, which translates designs directly into production-ready code. The features aim to automate routine tasks while enhancing designer creativity and productivity.

"Our AI strategy isn't about replacing designers, but supercharging their capabilities. These tools handle the tedious aspects of design work so that designers can focus on creative problem-solving".

Dylan Field, Figma CEO

Business & Regulation

US Government Scrutinizes Benchmark's Investment in Manus AI

The Committee on Foreign Investment in the United States (CFIUS) has initiated a review of Benchmark Capital's recent investment in Manus AI, a startup specializing in advanced robotics and machine learning systems. The review focuses on potential national security implications, given Manus AI's work on technologies with possible dual-use applications. This scrutiny reflects the increasing government oversight of AI investments, particularly those involving technologies with strategic significance.

Manus AI

"This review highlights the evolving regulatory landscape for AI investments. Investors and startups alike need to be prepared for increased scrutiny, especially in areas touching on national security concerns".

Patricia Hughes, Regulatory Expert

Conclusion

This week's developments reveal a fascinating dichotomy in AI: unprecedented collaboration through protocol standardization alongside intensifying competition through corporate policies and specialized models. The Microsoft-Google protocol partnership signals a maturing industry recognizing the value of interoperability, while Microsoft's restrictions on DeepSeek usage highlight growing data security concerns around AI tools. Docker's beta launch of the MCP Catalog and Toolkit represents a significant step toward standardizing agentic AI development, potentially doing for AI what containers did for application deployment. Mistral's Medium 3 demonstrates how specialized, efficient models are becoming competitive with larger offerings from established players, while Cisco's open-source security model represents a shift toward community-driven AI development in critical domains. Figma's design innovations show how AI is transforming creative workflows, while the Treasury Department's examination of Benchmark's Manus AI investment reveals the complex geopolitical dimensions of AI development. Looking ahead, we can expect both standardization and specialization trends to accelerate, creating a more interconnected yet diverse AI ecosystem that operates within evolving regulatory frameworks.

The Author

Luis Poveda

Luis Poveda’s AI Newsletter

Luis Poveda is a technology optimist and passionate innovator, constantly exploring and researching the latest trends. Based in Barcelona, he is currently focused on AI and developing a modern AI-driven IT network observability tool. He is also the creator and maintainer of Luis Poveda's AI Newsletter, where he curates and shares key insights on the evolving AI landscape.

Unlock the Secrets of Science, Philosophy, and Culture.

Discover stories that stretch your mind and stir your imagination. Nautilus, the science newsletter, explores the deep connections between science, philosophy, and culture — one idea at a time. Join thousands of curious readers and get beautifully crafted essays delivered to your inbox every week.