PLUS: Browsers that build their own UI, Google's Gemma 4 running offline, and China's OpenClaw agent
Good morning
A compelling new argument suggests the wait for AGI is over—because it's already here. The concept redefines AGI not as a single, future super-model, but as a system created by combining today's powerful LLMs with external "scaffolding" like tool-calling and autonomous agents.
This perspective shifts the entire focus of AI development from waiting for the next big model to innovating on the infrastructure around the ones we already have. If AGI is a system we can assemble right now, the question becomes: how do we best leverage these systems and what new opportunities does this unlock?
In today’s Next in AI:
The compelling case for why AGI is already here
Browsers that build their own personalized UI
Google's Gemma 4 running completely offline
China's 'lobster' AI agent craze
The AGI Debate is Over
Next in AI: A compelling new argument suggests AGI is not a future milestone but a present reality, achieved by combining today's powerful LLMs with "scaffolding" like tool-calling and autonomous agents.
Explained:
- The core idea is that AGI emerges from a system, not a single model. This scaffolding—an ecosystem of tools, protocols, and frameworks—allows an LLM to interact with the world, access information, and even learn new skills beyond its initial training.
- This combined system already meets many proposed definitions of AGI, from solving unfamiliar tasks and demonstrating creativity to performing economically valuable work without direct human oversight.
- This perspective shifts the focus of AI progress from simply waiting for the next model release to innovating on the infrastructure around them, with autonomous agentic systems that can operate 24/7 representing the new frontier.
Why It Matters: This argument reframes the conversation from "when will AGI arrive?" to "how do we best build with the AGI systems we already have?" It signals a major opportunity for developers and entrepreneurs to create value by building better tools and infrastructure on top of existing powerful models.
The End of Frontend?
Next in AI: A radical proposal imagines browsers that build their own UI, generating a personalized interface from a service's "manifest" and your unique preferences for layout and design. You can check out a proof-of-concept adaptive browser that demonstrates this idea.
Explained:
- The current web development model creates massive duplicated effort, with every service hand-crafting its own frontend, accessibility layer, and theme system from scratch.
- While concepts like Server-Driven UI and Generative UI frameworks exist, they keep control on the server side, with the service ultimately dictating the final user experience.
- This new model flips the script by having services publish a semantic manifest—like an API spec—that your browser uses to generate a fully bespoke interface based on your universal preferences.
Why It Matters:
This approach would make true personalization and accessibility the default for every application, baked directly into the browser. For businesses, the competitive advantage shifts from designing a unique frontend to building the most powerful and well-documented APIs.
Google's On-Device AI Sandbox
Next in AI: Google has released AI Edge Gallery, an open-source app that puts a powerful AI sandbox in your pocket, letting you run models like the new Gemma 4 completely offline.
Explained:
- The app's centerpiece is official support for the new Gemma 4 family of models, which brings advanced reasoning and creative capabilities directly to your hardware without an internet connection.
- Beyond chat, the app lets you augment the AI with "Agent Skills," allowing it to connect to tools like Wikipedia for fact-checking or even perform offline device controls and automated tasks.
- For developers and AI enthusiasts, a "Thinking Mode" reveals the model's step-by-step reasoning, while a Prompt Lab provides a dedicated workspace for testing prompts and model parameters.
Why It Matters: This move democratizes access to high-performance AI, allowing anyone to experiment safely on their own hardware with total data privacy. It also signals a major push toward building a community-driven ecosystem for on-device AI agents that can perform real-world tasks.
China's AI 'Lobster' Craze
Next in AI: An open-source AI agent named OpenClaw has sparked a nationwide frenzy in China. Users are creatively customizing the tool, nicknamed the "lobster," for everything from running e-commerce shops to trading stocks.
Explained:
- OpenClaw’s open-source code lets users connect it to local Chinese AI models. This is a key advantage because most Western tools, including ChatGPT, remain inaccessible in the country.
- The trend is fueled by Beijing's "AI Plus" national strategy, with cities offering millions in subsidies for adoption. Major tech companies are responding, with Tencent launching its own enterprise-grade platform, Tencent's ClawPro.
- This AI craze also addresses a major economic challenge by promoting AI-powered one-person companies to help curb youth unemployment, creating intense social pressure to learn and use the new technology.
Why It Matters: This phenomenon represents one of the largest real-world stress tests of agentic AI, driven by both government policy and grassroots adoption. It also signals China's aggressive strategy to embed AI throughout its economy and find practical applications to gain a competitive edge.
AI Pulse
An analysis details how AI war-game simulations failed to account for Iran's "mosaic defense" architecture, a key factor in the botched decapitation strike during the fictional Operation Epic Fury.
Anthropic standardized tool-calling with its Model Context Protocol (MCP), creating a common interface for models to connect with any service and accelerating the development of agentic systems.
Google shipped its Natively Adaptive Interfaces framework, which uses AI agents to make accessibility a default feature within an app's boundaries as a step toward more user-centric UI.