Google is rolling out its latest model, Gemini 3.5 Flash-Lite, within Google Search to support agentic search experiences and potentially enhance features like AI Overviews and AI Mode. This new model focuses on providing low latency and high throughput for workflows such as document processing and multi-step subagent tasks.
* Improved instruction following and better understanding of user intent
* Significant performance gains in coding (Terminal-Bench) and real-world task execution
* Optimized for agentic systems through built-in computer use tools and high-volume scalability
* Improved instruction following and better understanding of user intent
* Significant performance gains in coding (Terminal-Bench) and real-world task execution
* Optimized for agentic systems through built-in computer use tools and high-volume scalability
This guide provides a comprehensive walkthrough on using Google's Gemma 4 model to build autonomous AI agents through tool calling. It explores how this feature enables models to move beyond simple text generation by interacting with external APIs and systems via structured function calls.
Key topics covered in the article include:
- The mechanics of the tool calling loop, from reasoning and selection to execution and final response.
- Setting up a Python development environment using Hugging Face and necessary libraries like transformers and torch.
- Defining JSON schemas for tools to ensure precise model understanding.
- Implementing a full agent workflow by parsing function call responses and executing Python functions.
- A practical end-to-end demonstration of building a weather lookup agent.
- Managing multi-turn conversations through state management and conversation history.
- Best practices for production deployment, including argument validation, execution timeouts, and logging.
Key topics covered in the article include:
- The mechanics of the tool calling loop, from reasoning and selection to execution and final response.
- Setting up a Python development environment using Hugging Face and necessary libraries like transformers and torch.
- Defining JSON schemas for tools to ensure precise model understanding.
- Implementing a full agent workflow by parsing function call responses and executing Python functions.
- A practical end-to-end demonstration of building a weather lookup agent.
- Managing multi-turn conversations through state management and conversation history.
- Best practices for production deployment, including argument validation, execution timeouts, and logging.
The article discusses the benefits of running Google’s Gemma 4 models locally on personal hardware to ensure data privacy and independence from cloud services. By executing these multimodal models on a laptop, users can process images and audio without exposing sensitive information to third-party servers. The text highlights how efficient architecture allows for vision capabilities and speech recognition even with limited VRAM.
* Localized processing avoids the risks of uploading private or regulated data
* Native audio support in specific model variants like E2B and E4B
* Large context windows allow for deep analysis of lengthy documents and codebases
* Reduced reliance on internet connectivity during mobile workflows
* Localized processing avoids the risks of uploading private or regulated data
* Native audio support in specific model variants like E2B and E4B
* Large context windows allow for deep analysis of lengthy documents and codebases
* Reduced reliance on internet connectivity during mobile workflows
Andrej Karpathy, Google, and Garry Tan are all adopting Markdown as the primary format for agent memory and organizational context. While Karpathy focuses on personal knowledge, Google targets enterprise runbooks, and Tan defines agent roles, they all share a reliance on versioned Markdown files. This shift suggests that the strategic advantage is moving from the specific large language model used to the curated knowledge base a team accumulates.
- Karpathy's LLM Wiki for personal knowledge bases
- Google's Open Knowledge Format for enterprise context
- Garry Tan's gstack for defining agent roles
- The migration of the competitive moat from models to portable data files
- Karpathy's LLM Wiki for personal knowledge bases
- Google's Open Knowledge Format for enterprise context
- Garry Tan's gstack for defining agent roles
- The migration of the competitive moat from models to portable data files
Google's release of Gemma 4 marks a major turning point for open-source AI, offering a versatile family of multimodal models under a permissive Apache 2.0 license. Built using Gemini 3 technology, these models demonstrate massive leaps in math and coding performance, rivaling much larger proprietary systems while remaining efficient enough to run on local hardware ranging from smartphones to high-end GPUs. This release positions Google as a formidable competitor in the open-weights ecosystem, prioritizing user ownership and deployment efficiency.
* Apache 2.0 license
* Multimodal intelligence
* Local hardware deployment
* Massive benchmark leaps
* Efficient MoE architecture
**Models**
* E2B: Mobile efficiency
* E4B: Edge specialist
* 26B MoE: Speed meets intelligence
* 31B Dense: Top-tier performance
* Apache 2.0 license
* Multimodal intelligence
* Local hardware deployment
* Massive benchmark leaps
* Efficient MoE architecture
**Models**
* E2B: Mobile efficiency
* E4B: Edge specialist
* 26B MoE: Speed meets intelligence
* 31B Dense: Top-tier performance
Google's web.dev guidance now advises developers to treat AI agents as a distinct audience alongside human visitors. As more users delegate goal-oriented tasks to AI, websites with complex hover states or shifting layouts may become functionally broken for these automated entities. The guide highlights that optimization for agents aligns closely with existing accessibility and semantic HTML best practices, making sites better for both humans and machines.
* Treating agents as a distinct visitor type
* How agents interpret websites via screenshots, raw HTML, and the accessibility tree
* Recommendations for using semantic HTML elements and maintaining stable layouts
* Introduction to WebMCP, a proposed web standard for agent-website interaction
* Treating agents as a distinct visitor type
* How agents interpret websites via screenshots, raw HTML, and the accessibility tree
* Recommendations for using semantic HTML elements and maintaining stable layouts
* Introduction to WebMCP, a proposed web standard for agent-website interaction
Mozilla is expressing strong opposition to Google's implementation of a Prompt API in the Chrome and Edge browsers, which allows web pages to interact directly with local machine learning models like Gemini Nano. The organization warns that this integration could undermine web interoperability and neutrality by forcing developers to optimize for specific vendor models and adhere to proprietary content policies.
Main points:
- Risk of creating model-specific code paths that harm browser compatibility.
- Concerns regarding the imposition of vendor-specific usage rules on an open platform.
- Disagreement over whether there is a genuine groundswell of developer support for the API.
Main points:
- Risk of creating model-specific code paths that harm browser compatibility.
- Concerns regarding the imposition of vendor-specific usage rules on an open platform.
- Disagreement over whether there is a genuine groundswell of developer support for the API.
Researchers from Google and Forcepoint have identified a rise in indirect prompt injection (IPI) attacks, where malicious instructions are hidden within web pages to manipulate LLM-powered AI agents. While some injections are harmless pranks or tone adjustments, others aim for serious harm including traffic hijacking, data exfiltration, denial of service, and financial fraud through unauthorized payment processing. Attackers use techniques like invisible text, HTML comments, and metadata manipulation to hide these payloads from humans while remaining visible to AI.
Key points:
* Real-world evidence of IPI attacks found in massive web crawls and active threat hunting.
* Malicious intents include search engine manipulation, data theft (API keys), and destructive commands.
* Financial fraud attempts have been observed using embedded PayPal transactions and Stripe donation routing.
* Attackers hide instructions via single-pixel text, near-transparent colors, or metadata injection.
* The risk level scales with AI privilege; agentic AIs capable of executing commands or payments are high-impact targets.
Key points:
* Real-world evidence of IPI attacks found in massive web crawls and active threat hunting.
* Malicious intents include search engine manipulation, data theft (API keys), and destructive commands.
* Financial fraud attempts have been observed using embedded PayPal transactions and Stripe donation routing.
* Attackers hide instructions via single-pixel text, near-transparent colors, or metadata injection.
* The risk level scales with AI privilege; agentic AIs capable of executing commands or payments are high-impact targets.
The author explores how Gemini Scheduled Actions represents a significant shift in Android automation by moving from rigid, trigger-based logic like Tasker to an intent-first architecture powered by Large Language Models. Unlike traditional tools that require programming knowledge and are prone to breaking when UI changes occur, Gemini understands natural language requests and manages complex workflows across devices via the cloud.
Key points:
* Comparison between brittle IFTTT engines and flexible LLM-based automation.
* The benefit of cross-device synchronization through Google accounts.
* Using the desktop web interface for easier setup and access to an Inspiration Gallery.
* Practical use cases including automated SEO idea generation, sports updates, grocery list creation in Google Keep, and email summaries.
* Current limitation of up to 10 active scheduled actions at a time.
Key points:
* Comparison between brittle IFTTT engines and flexible LLM-based automation.
* The benefit of cross-device synchronization through Google accounts.
* Using the desktop web interface for easier setup and access to an Inspiration Gallery.
* Practical use cases including automated SEO idea generation, sports updates, grocery list creation in Google Keep, and email summaries.
* Current limitation of up to 10 active scheduled actions at a time.
A Python package designed to provide production-ready templates for Generative AI agents on Google Cloud. It allows developers to focus on agent logic by automating the surrounding infrastructure, including CI/CD pipelines, observability, security, and deployment via Cloud Run or Agent Engine.
Key features and offerings include:
- Pre-built agent templates such as ReAct, RAG (Retrieval-Augmented Generation), multi-agent systems, and real-time multimodal agents using Gemini.
- Automated CI/CD integration with Google Cloud Build and GitHub Actions.
- Data pipelines for RAG using Terraform, supporting Vertex AI Search and Vector Search.
- Support for various frameworks including Google's Agent Development Kit (ADK) and LangGraph.
- Integration with the Gemini CLI for architectural guidance directly in the terminal.
Key features and offerings include:
- Pre-built agent templates such as ReAct, RAG (Retrieval-Augmented Generation), multi-agent systems, and real-time multimodal agents using Gemini.
- Automated CI/CD integration with Google Cloud Build and GitHub Actions.
- Data pipelines for RAG using Terraform, supporting Vertex AI Search and Vector Search.
- Support for various frameworks including Google's Agent Development Kit (ADK) and LangGraph.
- Integration with the Gemini CLI for architectural guidance directly in the terminal.
