Google's enterprise agent platform is here

Inclusion AI

LLaDA2.0-Uni introduces a unified discrete diffusion LLM that natively supports both multimodal understanding and generation, matching specialized models in comprehension while enabling strong image generation and editing capabilities within a single scalable architecture.

Microsoft

Microsoft's AutoAdapt automates the complex process of adapting LLMs to specialized domains by planning optimal strategies (RAG vs fine-tuning), selecting configurations, and tuning hyperparameters within real-world constraints like cost and latency. The open-source framework transforms weeks of manual iteration into reproducible pipelines, crucial for high-stakes applications in healthcare, law, and enterprise operations.

Google Cloud Blog

Google Cloud launched Gemini Enterprise Agent Platform, consolidating Vertex AI into a comprehensive system for building, deploying, and governing enterprise AI agents at scale. The platform features multi-day agent persistence, memory capabilities, centralized security controls, and support for 200+ models, with customers like Comcast and L'Oréal already deploying production agents.

Google DeepMind

Google DeepMind's Decoupled DiLoCo enables resilient distributed AI training across distant data centers with 200x less bandwidth, maintaining 88% training efficiency even with high hardware failure rates while allowing mixed-generation hardware in single training runs.

DeepSeek's January mHC paper reveals training and inference optimizations that have been incorporated into their V4 model, showcasing the company's rapid iteration on architectural improvements.