Global AI Weekly - Issue 134 - Global AI Community

In this issue

Highlights

3 stories

axios.com

OpenAI Makes Its Hardware Move in 2026

OpenAI is moving beyond software with plans to unveil its first AI hardware device in the second half of 2026, Axios reports. The gadget, whose design direction is being influenced by famed former Apple designer Jony Ive, is expected to be compact and innovative, though specific features haven’t been disclosed yet. OpenAI’s chief global affairs officer says more details will come later this year as the company pushes into new user experiences beyond screens.

theverge.com

Claude Code is suddenly everywhere inside Microsoft

Microsoft is now exploring the use of Claude Code within its teams, even though it already offers its own AI tool, GitHub Copilot. Developers at Microsoft are actively testing both AI tools to evaluate their effectiveness. This move highlights Microsoft's openness to experimenting with multiple AI solutions to enhance productivity.

testingcatalog.com

Anthropic works on Knowledge Bases for Claude Cowork

In this issue

Research

1 story

huggingface.co

Think-Then-Generate: Reasoning-Aware Text-to-Image Diffusion with LLM Encoders

This content explores the concept of improving text-to-image diffusion models by incorporating reasoning-aware techniques using Large Language Models (LLM) as encoders. It introduces the "Think-Then-Generate" approach, focusing on generating more coherent and contextually appropriate images by enhancing the model's understanding of textual input. The discussion emphasizes the integration of advanced reasoning capabilities to bridge gaps in current text-to-image generation methods.

In this issue

Video

2 stories

youtube.com

Designing a Flexible Evaluation Backend for LLM Applications

This talk introduces Rhesis, an open-source evaluation backend designed to unify diverse frameworks like DeepEval, Ragas, and Promptfoo into a single, extensible interface for assessing large language models (LLMs). It covers the engineering strategies, design patterns, and abstractions employed to create a modular and adaptable system that supports current and future integrations. With expertise from Nicolai Bohn and Arkadiusz Kwasigroch, participants gain insights into testing, collaboration tools, and practical approaches for building robust evaluation backends in their own AI projects.

youtube.com

Federated Learning & Encrypted AI Agents: Secure Data & AI Made Simple

Prachi Modi explores Federated Learning and Encrypted AI Agents, innovative technologies ensuring data privacy while training AI models. These advancements use techniques like homomorphic encryption and secure aggregation to create ethical and secure AI systems. Learn how these tools are shaping the future of AI with a focus on data security and privacy protection.

In this issue

Articles

4 stories

marktechpost.com

Liquid AI Releases LFM2.5-1.2B-Thinking: a 1.2B Parameter Reasoning Model That Fits Under 1 GB On-Device

Liquid AI has introduced LFM2.5-1.2B-Thinking, a compact 1.2 billion parameter reasoning model designed to operate efficiently on devices, requiring less than 1 GB of storage. As part of the LFM2.5 family, this model aims to bring advanced reasoning capabilities to on-device applications without compromising performance or accessibility. It represents a step forward in making powerful AI tools more portable and resource-efficient.

infoworld.com

Visual Studio Code adds agent development extension

The Copilot Studio extension for Visual Studio Code allows developers to create AI agents using any AI assistant compatible with VS Code. It ensures smooth integration with Copilot Studio for seamless testing and iteration, streamlining the AI development process. This tool aims to enhance productivity and simplify workflows for developers working on AI projects.

openai.com

Unlocking Your AI Superpower: From Capability to Personal Empowerment

towardsdatascience.com

Using Local LLMs to Discover High-Performance Algorithms

In this issue

Upcoming Events

1 story

globalai.community

AgentCon - The AI Agents World Tour Continues in 2026

AgentCon continues into 2026 with the AI Agents World Tour—one-day, developer-focused conferences dedicated to autonomous AI agents. Building on a successful run of events, the tour expands to even more cities worldwide, from San Francisco to Singapore and beyond. Join leading engineers, researchers, and builders to explore cutting-edge agent architectures, real-world use cases, and emerging best practices. Connect with the global AI community and help shape the future of autonomous AI.

In this issue

Code

2 stories

cursor.com

Scaling long-running autonomous coding

We’ve been testing the potential of autonomous coding agents, running them continuously for weeks to explore their long-term capabilities. This experimentation highlights how these agents tackle complex coding challenges over extended periods. The focus is on understanding their scalability and efficiency in managing persistent tasks.

philschmid.de

MCP is Not the Problem, It's your Server: Best Practices for Building MCP Servers

The hype around the Model Context Protocol (MCP) led to a surge in MCP server development, but many have fallen short of expectations. Rather than blaming the protocol itself, the real issue often lies in poor server implementation. This piece focuses on addressing common pitfalls and offers best practices for building better MCP servers to ensure better performance and user satisfaction.

In this issue

Podcast

1 story

artificialintelligencepod.com

Artificial Intelligence Podcast

The Artificial Intelligence Podcast offers engaging conversations with experts in AI, machine learning, and related fields. The episodes explore the technology, its advancements, and the ways it is shaping the world. Each discussion provides insights into the impact of AI on society and its future possibilities.