September 2, 2026
Quick hits from across the AI world in the last 24 hours.
- ⚡ Quick hits
- Agents MIT Technology Review published guidance on scaling enterprise agentic-AI pilots beyond isolated experiments and into production operations. MIT Technology Review ↗
- Agents Cursor introduced Origin, an early-beta Git forge where agents and teams can create repositories, mirror GitHub projects, and manage pull requests. InfoQ 中文 ↗
- Industry Anthropic signed a $35 billion infrastructure agreement with Lambda to expand the computing capacity supporting Claude. The Decoder ↗
- Policy Sakana AI discussed resilience and lessons from nuclear deterrence as foundations for governing AI used in defense. X / Sakana AI ↗
- Industry Meshy and SHAKE4 announced a free, approval-based Tokyo meetup connecting game-industry participants after Tokyo Game Show. X / Meshy ↗
- Agents Genspark added Gemini 3.8 Flash across AI Chat, Code Agent, and Claw immediately after the model became available. X / Genspark ↗
- AIGC Hailuo AI published a generative-video demonstration framing an ordinary object with a newly created behind-the-scenes sequence. X / Hailuo AI ↗
- AIGC Wan 3.0 debuted first for video editing and second for text-to-video with audio on Artificial Analysis’s leaderboards. X / Artificial Analysis ↗
- Models Baseten released GLM-5.3 Fast, positioning the open-weight model for latency-sensitive applications requiring higher token throughput. X / Baseten ↗
- Agents Replit enabled project analytics for all new projects and lets its agent suggest application-specific custom metrics. X / Replit ↗
- Research Hugging Face demonstrated improving structured output from a 350-million-parameter model using 100 GRPO training steps. Hugging Face ↗
- Agents Hugging Face introduced Funes, an open approach for giving coding agents persistent memory controlled by users. Hugging Face ↗
- AIGC Hugging Face trained a coding model with TRL and OpenEnv to generate programs that render watercolor-style images. Hugging Face ↗
- Open Source Qwen developers open-sourced zg, a local-first workspace search tool combining ripgrep, BM25, vector retrieval, and agent integrations. GitHub ↗
- Agents GitHub added Copilot integrations for Slack and Microsoft Teams, new models, and a Balanced code-review depth. GitHub on X ↗
- Agents Claude can now operate desktop applications in the background through Claude Cowork and Claude Code on macOS. Claude on X ↗
- Agents VS Code previewed Agent Merge, which addresses review comments, failed checks, and merge conflicts before pull requests land. Visual Studio Code on X ↗
- Industry The Open Source AI Alliance began transitioning to the Linux Foundation to advance open, practical, and secure AI development. NVIDIA AI on X ↗
- Open Source GitHub Next open-sourced Chopin, a real-time multiplayer environment for teams to plan agent work collaboratively. GitHub Next on X ↗
- Policy The US Justice Department backed OpenAI’s fair-use argument for training models on copyrighted material in its dispute with The New York Times. TechCrunch ↗
- AIGC Adobe Firefly introduced an AI video upscaler designed to sharpen low-resolution footage, including older personal videos. Adobe Firefly on X ↗
- Funding Enterprise AI automation startup Wonderful raised $550 million at a $5 billion valuation, more than doubling its valuation within six months. TechCrunch ↗
- Research LlamaIndex released ExtractBench on Kaggle to test schema-guided extraction from documents likely to break downstream agent workflows. LlamaIndex on X ↗
- AIGC Runway introduced Dev MCP for building, debugging, and managing its developer platform directly from compatible coding agents. Runway on X ↗
- Open Source Unsloth added MTP support for local Qwen3.8-Flash inference, claiming 1.3–1.7× speedups and up to 170 tokens per second on specified hardware. Unsloth AI on X ↗
- Industry ElevenLabs appointed former GitLab executive Ashley Kramer as chief revenue officer to lead global customer deployment and commercial growth. ElevenLabs on X ↗
- Research AI21 Labs trained lower-cost independent verifiers to improve agent answers, reporting results on Google DeepMind’s FACTS-Search benchmark. AI21 Labs on X ↗
- Safety ZeroDrift launched a service that checks AI-agent-generated customer messages against an organization’s policies before delivery. SiliconANGLE ↗