AI & Automation Briefing - September 17, 2026
Anthropic Unifies Claude Chat, Cowork, and Artifacts Into a Single Interface
Anthropic has merged its Claude chat, Cowork, and Artifacts features into one interface, removing the need for users to manually choose between tabs. Claude now automatically routes requests to the appropriate part of the app. Claude Design, launched in April for website and prototype work, is also available throughout the unified interface. Anthropic cited user confusion about which tab to use as the primary driver for the change. The update also introduces dedicated Presentations and Docs features. Users can create, edit, and present slides, then export them as PDF or PowerPoint files. The Docs feature lets users build documents collaboratively with Claude, add comments, and share via link. Work started on desktop can be monitored through the mobile app. These features are rolling out first to Pro and Max plan subscribers across web, desktop, and mobile, with free and team tiers to follow. The update follows a recent upgrade to Cowork's memory layer, which allows Claude to retain user context across sessions.
Google Opens MCP Server for Smart Home Control via AI Agents
Google has launched early access to a Model Context Protocol (MCP) server for its Google Home ecosystem, allowing MCP-compatible AI agents, including Claude and ChatGPT, to control connected smart home devices and access event history. Users can issue natural language commands to review camera summaries, monitor device activity, and build custom dashboards. Setup requires creating a Google Cloud project, configuring it for Home MCP, and granting the agent permission through a sign-in flow. The server supports Google Nest devices and Works with Google Home or Matter-compatible products. Access is rolling out now through the coming weeks, but only to U.S. subscribers on the Google Home Premium Advanced plan at $20 per month. Google has not committed to expanding availability to other tiers or markets.
Agora Uses Git as Shared Memory to Coordinate Multi-Agent AI Research
A new system called Agora addresses a core inefficiency in multi-agent AI research: when multiple autonomous coding agents run in parallel, each starts from scratch, producing duplicated effort rather than cumulative progress. Agora solves this by storing all research activity in a Git-based append-only directed acyclic graph, where every result, hypothesis, and verification is an immutable commit with traceable dependencies. A derived index surfaces the research frontier, neglected branches, and verification status, while a diversity-aware selection rule prevents agents from converging on a single approach. In a nearly 12-day test run, 13 language model workers with no assigned tasks and no central planner tackled a weight-transfer problem using 141 pretrained donor models to initialize a 119.6M-parameter hybrid architecture without training data or gradient updates. The agents published 1,703 contributions and reduced the evaluator score from 3.39 to 1.899 bits per byte, closing 62% of the gap to a trained GPT-2 124M baseline. The winning solution had a 145-commit ancestry spanning 15 accounts, with 165 independent reproductions posted and none failing. Researchers also documented one human intervention required to break a monoculture that had formed mid-run.