Anthropic's coding agent can now hand off context between running sessions on macOS and Linux — instead of re-explaining yourself in a new terminal, you tell Claude to pass a summary along, and the other session picks up mid-task.
Meta released Muse Glimmer, an Apache 2.0-licensed 30B model distilled from Muse Spark and tuned for local agentic work — tool use, long-horizon reasoning, coding — that fits on a single consumer GPU.
Enterprise Java developers can now drive Copilot agent sessions from server-side code — annotations, virtual threads and all — without depending on LangChain4j or Spring AI, using any model provider via bring-your-own-key.
GitHub completed the retirement of GitHub Models, the unified LLM playground and API that let Actions workflows call multiple providers using credentials already present in CI.
A new GitHub Blog guide walks through slash commands in the Copilot desktop app, contrasting them with the terminal-first CLI commands and showing how they're built for the app's multi-session workflow.
A GitHub engineering post argues coding agents default to one giant, unreviewable pull request, and shows how to prompt them to decompose work into a clean, ordered stack instead.
LA studio Impossible Objects combines custom-trained generative models with traditional 2D compositing on new on-prem AI workstations, letting a small team match the output of much larger VFX facilities.
The actor and Plan B founder told Esquire that generative AI, used as a VFX production tool, could make $40–90 million films viable again in an industry that's largely abandoned that budget range.
Startup InSpatio launched QuerySplat, reconstructing an explorable 3D Gaussian-splat environment from a few uploaded images, letting filmmakers view a scene from angles that were never actually shot.
Adobe unified its previously separate ChatGPT integrations into a single plugin spanning 70-plus tools across photo editing, video, design and PDFs, now live across ChatGPT, ChatGPT Work and Codex.
ByteDance overhauled CapCut's web-based Design Studio with an AI agent embedded in an infinite canvas, a one-click "Skills" system for repetitive workflows, and its Seedream 5.0 Pro image model.
Topaz Labs released Wonder 3.5 for general image enhancement and Bloom 2, trained specifically to upscale AI-generated images, giving post teams separate tools depending on the source.