Nvidia has expanded its open-source agentic AI portfolio with the Nemotron 3 series, a family of Mixture-of-Experts (MoE) models that decouple size from compute cost to deliver up to four times the performance of previous generations. Nvidia is releasing the models alongside trillions of training tokens, reinforcement learning libraries, and new “gym” environments. TNS Senior Editor for AI Frederic Lardinois looks at the offering and why Nvidia builds its own models.

In other technology developments we’re covering:

• Fedora Silverblue is an ideal Linux distribution for developers thanks to Toolbox, a preinstalled utility that creates mutable, containerized environments for coding and debugging without cluttering the stable, immutable host OS. Technology reporter Jack Wallen explains how the containers allow developers to safely experiment with different distributions and dependencies. Jack’s tutorial walks you through creating a new Toolbox environment.
• Web development in 2025 was defined by a push toward simplicity through native web features and the “Vite-ification” of the frontend ecosystem, writes TNS Senior Editor Richard MacManus. This happened even as AI coding assistants created a counter-trend by defaulting to React-heavy architectures. Richard writes that, ultimately, the industry remains at a crossroads where the growing reliance on AI tools threatens to overshadow efforts to use lightweight, vanilla JavaScript solutions.

Google's Agent Development Kit (ADK) is a fundamental shift in how developers build AI-powered applications. Rather than treating large language models as simple request-response systems, ADK introduces an event-driven runtime architecture that orchestrates agents, tools and persistent state into cohesive applications. This article explores the core...

Cloud outages keep proving the same point: a snapshot won’t bring your stack back to life. What matters is whether your cloud can be rebuilt. Firefly finds the drift, the gaps, and the hidden dependencies—so your team can actually recover when it counts.

Most AI initiatives never reach production. Learn what’s really blocking enterprise AI success and get actionable solutions from Forrester and Anaconda experts.