Loading...
Loading...
Google quietly introduced WebMCP at I/O 2026: an open web standard prototype in Chrome 149 that lets developers annotate JavaScript functions and HTML forms so browser-based AI agents can call them as typed, structured tools. Instead of brittle DOM scraping or bespoke API connectors, WebMCP exposes a manifest of callable actions (e.g., schedule_demo, generate_proposal) with defined parameters and handlers, enabling reliable agent-to-website interactions. For builders of conversational UIs and ag
WebMCP in Chrome 149 creates a standardized way for browser-based AI agents to call site-provided functions and forms, reducing fragile scraping and bespoke connectors. Tech professionals building conversational UIs, agents, or site integrations gain a more reliable channel for agent-to-website interactions and tool invocation.
Dossier last updated: 2026-05-31 15:29:42
A developer reflects on how AI tooling supercharged rapid project creation—everything from speech recognition in Rust, an Invidious clone, a Jellyfin-like desktop app, to a regional news site and a SaaS—but left them with many unfinished, unmaintainable projects and rising cognitive costs. They argue large language models amplify distraction and fragmentation: tooling encourages quick output, token usage and endless follow-ups rather than disciplined, maintainable engineering. Attempts to limit usage with quotas or alternate models (Claude, Codex) only partially helped. The author warns that current AI interfaces prioritize engagement and token consumption over focused, product-driven work, and suggests friction is necessary to preserve attention and sustainable development.
Google quietly introduced WebMCP at I/O 2026: an open web standard prototype in Chrome 149 that lets developers annotate JavaScript functions and HTML forms so browser-based AI agents can call them as typed, structured tools. Instead of brittle DOM scraping or bespoke API connectors, WebMCP exposes a manifest of callable actions (e.g., schedule_demo, generate_proposal) with defined parameters and handlers, enabling reliable agent-to-website interactions. For builders of conversational UIs and agentic workflows, this could make every website a first-class tool surface, reducing automation fragility and integration overhead. If widely adopted, WebMCP could shift how AI agents execute tasks on the web and accelerate agent-driven web apps.
A V2EX forum post humorously compares major AI chatbots to anime-style characters, labeling Codex, Claude and Gemini as the “御三家” and mapping personality traits: Codex as introverted and widely used in China; Claude as calculating, strict and prone to banning users; Gemini as carefree and emotionally engaging but limited. The thread frames these descriptions as playful metaphors rather than technical analysis, notes a mention of MyGO (五人) and prompts commenters to ask about Grok and Gemini. It matters as a cultural snapshot of how Chinese-speaking tech communities personify competing AI models, reflecting user perceptions and brand image rather than objective capabilities.
Major commercial models like Google’s Gemini and Anthropic’s Claude Code are effectively throttled or simplified because running them at full capability requires vast compute and specialized hardware; individual PCs and idle GPUs can’t practically match those requirements. These models use large parameter counts, optimized runtimes, and server-scale infrastructure (multi-GPU clusters, high-memory accelerators, and fast interconnects) plus proprietary optimizations and licensing that prevent simple local deployment. The result: smaller, distilled, or cost-cut models for everyday use, and cloud-hosted inference to manage latency, updates, safety controls, and monetization. This matters because it shapes which AI capabilities remain centralized vs. locally accessible, affecting developer flexibility, costs, and privacy trade-offs.