- Published on
Claude Projects v2, Google family agent
- Authors
- Name
- DeepHorus
On September 18, 2026, AI labs unveiled agentic tools and research milestones. Anthropic launched Claude Projects v2, Google introduced a family cloud agent, and new benchmarks showed AI doing more research. These moves signal a shift from chatbots to autonomous systems.
Claude Projects v2 turns folders into conversations. It automates task delegation, coordination, and result assembly across cloud sessions, using threads for parallel work and shared memory. The beta is available to select subscribers, with wider access coming.
Google's family agent runs on its own cloud computer and Google account. It turns shared emails, files, and calendars into daily briefings, fills forms, and coordinates activities for up to six people. The US experiment is on a waitlist.
Anthropic says Claude now leads 26 percent of its AI research work and oversees tens of thousands of internal agents. It also open-sourced biomolecular modeling improvements: over 30 models optimized, a 4x speed gain, and a low-memory mode for larger predictions on one NVIDIA GPU. A protein design contest with Adaptyv Bio offers up to $1 million in Claude credits.
Z.ai used a GLM-5.3-powered Infra Agent to build GLM-5.3-Flash's serving stack on 100,000+ Chinese accelerators in under two weeks. Kernel fixes and dense feedback tripled throughput while humans kept control of objectives and risk. This shows agents helping build their own infrastructure.
OpenAI's Noam Brown discussed multi-agent systems, alignment, and recursive self-improvement in a new interview. Another analysis argues LLM classification is really feature engineering. These ideas are shaping how researchers approach safe scaling.
The developments raise oversight questions as AI contributes to its own development. Expect wider access to Claude Projects and Google's family agent soon. The industry is racing to balance capability with control.
