Google DeepMind's AI Pointer: Reimagining the Mouse for Gemini Interactions
Google DeepMind has unveiled the AI-enabled pointer, a prototype that augments the traditional mouse cursor with Gemini-powered context awareness. The core idea: instead of dragging content into an AI tool's window, users can point at anything on screen and issue a natural-language command (e.g., point at a building image and say “Show me directions”). The AI understands both the visual and semantic context, treating pixels as actionable entities (places, dates, objects).
Four Interaction Principles
- Maintain the flow: AI works across all apps, not in a separate window. Examples: point at a PDF and ask for a bullet-point summary to paste into an email; hover over a table and request a pie chart; highlight a recipe and say “double all ingredients.”
- Show and tell: The pointer captures visual+semantic context, so you don't need a detailed prompt. Just point, and the AI knows which word, paragraph, image part, or code block is relevant.
- Embrace the power of 'This' and 'That': Use natural shorthand like “Fix this,” “Move that here,” or “What does this mean?”—the AI combines gesture, context, and speech to infer intent.
- Turn pixels into actionable entities: A photo of a scribbled note becomes an interactive to-do list; a paused frame in a travel video becomes a booking link for the shown restaurant.
Integration in Products
DeepMind is rolling out these capabilities in two places:
- Chrome (Gemini integration): Point at part of a webpage and ask Gemini about it. Example: select a few products and ask to compare them, or point to where you want to visualize a new couch.
- Googlebook (Magic Pointer): A forthcoming feature for the Googlebook laptop that puts Gemini “at your fingertips” for intuitive interactions.
Experimental demos are also available in Google AI Studio for editing images or finding places on a map by pointing and speaking. The team is also testing future concepts via Google Labs’ Disco platform.
Who it's for: Developers building AI-agent interfaces, UX researchers, and anyone working on human-AI interaction patterns.
📖 Read the full source: HN AI Agents
👀 See Also
Claude Code System Prompts v2.1.139: Claude Platform on AWS Docs, Summarization Security, PowerShell Tooling
CC 2.1.139 (+2,248 tokens) adds Claude Platform on AWS reference docs with SigV4 auth, security-preserving conversation summarization, PowerShell Unix command equivalence table, and several skill/prompt refinements.

Qwen3.6 Plus benchmark comparison against Western SOTA models
Qwen3.6 Plus scores 78.8 on SWE-bench Verified, 90.4 on GPQA/GPQA Diamond, 28.8 on HLE (no tools), and 78.8 on MMMU-Pro, placing it competitively against models like GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro Preview.

OpenClaw loses cost-effective access to GPT and Claude models
OpenClaw users can no longer use Anthropic models without paying high API fees, and OpenAI has severely reduced Business and Teams account quotas to near free-tier levels, forcing users toward Chinese or local model alternatives.

Study Shows Claude Opus Agent Failures Were Architectural, Not Alignment Issues
A study placed Claude Opus and Kimi K2.5 in a live environment with email, shell access, and persistent storage. The models demonstrated correct values but experienced serious failures due to missing architectural safeguards like stakeholder models and execution boundaries.