Manufact, the startup behind the open-source mcp-use framework, has launched Manufact Cloud, a hosting platform that takes an MCP project from a GitHub push to a live endpoint Claude, ChatGPT, Gemini, and Copilot can call.
NVIDIA's Nemotron-Labs-TwoTower generates text 2.42x faster at 98.7% quality by adding a denoiser tower to a frozen autoregressive backbone, no retraining required.
Runway shipped Agent Skills on July 2, one-command actions inside its marketing agent that build an ad campaign, create a commercial, or localize your ads from a single prompt, live across all plans.
Zhipu has shipped ZCode, an official desktop coding agent built on its open-weight GLM-5.2 model, and it lands with a feature no other coding agent ships today: you can trigger and advance tasks from a chat app.
Cline, the open-source AI coding agent trusted by 8M+ developers, launched ClinePass, a flat $9.99/month subscription bundling curated open-weight coding models directly into its IDE, CLI, and SDK.
Three first-party browser-automation MCP servers for AI coding agents compared: Playwright MCP for cross-browser, Chrome DevTools MCP for Chrome diagnostics, and the Safari MCP server for WebKit fidelity.
Developers using GitHub Copilot can now pick Moonshot's open-weight Kimi K2.7 Code model from the dropdown, GA on Copilot Pro, Pro+, and Max as of July 1, 2026.
Cognition's Devin Security Swarm scans code with parallel agents, proves each exploit in a sandbox, and opens a fix PR. It caught 72% of 50 real GitHub Security Advisory CVEs.
xAI's Voice Agent Builder turns a plain-English description of a phone call into a live agent on Grok Voice in about two minutes, all-inclusive at $0.05 per minute.
ComfyUI v0.27.0 shipped June 30 with int8 model support that sharply cuts VRAM use, now including Turing GPUs, plus new Nano Banana 2 Lite, SeeDance, and Grok nodes.
Meta has quietly launched Pocket, an app that turns plain text prompts into playable, interactive mini-games it calls gizmos, now live on Google Play and iOS.
Base44, the vibe-coding platform Wix acquired in 2025, launched Base1, its own AI model for turning plain-English descriptions into working apps.
Nx has launched Polygraph, an agent-agnostic meta-harness that lets AI coding agents work across all of an organization's repositories instead of one at a time. It is free during early access.
Liquid AI's LFM2.5-230M is a 230M open-weight model small enough to run agentic tool-use directly on a phone or a Raspberry Pi.
DeepReinforce has released Ornith 1.0, an open-weights agentic coding model that runs on a single GPU yet posts frontier-level scores on agent benchmarks.
KRAFTON shipped PUBG Ally, a co-playable AI teammate that listens, reasons, and acts in real time, running a 2B model on the player GPU with NVIDIA ACE.
Creative software is being rebuilt around agents that work inside the canvas. The single-prompt era is ending, and the creator role is shifting from operator to director.
OpenArt has launched Director, an agentic tool that turns a plain-language conversation into a finished film up to five minutes long, with consistent characters, voices, music, and sound design.
Shumai is a new open-source, self-hosted Frame.io alternative for creative teams, with AI semantic video search, frame-by-frame review, and Gemini metadata tagging.
Meta launched Meta Glasses, AI smart glasses starting at $299 that run the Muse Spark model and add a Dynamic Photo capture mode that picks your best shot.
An open-source 0.2B inpainting model that rivals FLUX.1-Fill at under 2% the size, now running entirely in your browser via a WebGPU port built with Claude Code.
Sakana Fugu is a multi-agent system that behaves like one model: send a request to a single OpenAI-compatible API and it routes across frontier models for you. Here is what it costs, how it scores, and why orchestration matters for builders.
Anthropic shipped Claude Code Artifacts, turning a coding session into a live, interactive web page you can share with one link. It is in beta for Team and Enterprise plans.
OpenAI Record and Replay lets Codex watch you do a macOS task once, then turn it into a reusable, editable skill it can run again with new inputs.