Blog
Articles about AI agent readiness, web standards, and preparing for autonomous agents.
WebMCP Is Live in ChatGPT and Codex: What Changed in August 2026
OpenAI turned on WebMCP site tools in the ChatGPT desktop browser and Codex on 25 August, three weeks after Shopify switched it on for every Liquid storefront. Three live deployments show the part nobody mentions: from the outside, you can't see any of it.
Does Schema Markup Get You Cited by AI? What the Data Actually Shows
A 2026 Ahrefs study tracked 1,885 pages adding schema markup, and AI-search citations barely moved. We separate correlation from causation: what structured data is actually proven to do, what really drives citations, and where to spend your effort.
The Responsive Design Moment for AI Agents
The web is going through the same shift it went through with mobile. First separate m.dot sites, then responsive convergence. AI agent readiness is following the same arc: separate agent APIs today, one adaptive content layer tomorrow. We trace the parallels, the data, and what comes after convergence — the personal LLM layer.
Content Negotiation for AI Agents: Why Sentry Serves Markdown Over HTML
Sentry co-founder David Cramer shows how content negotiation — a 25-year-old HTTP standard — saves AI agents 80% of tokens. We break down the implementation: Accept headers, markdown delivery, authenticated page redirects, and what this means for every website preparing for agent traffic.
Vercel's agent-browser: Why a CLI Beats MCP for Browser Automation
Vercel's agent-browser hit 22,000 GitHub stars in two months. It's a CLI, not an MCP server, and the data shows why: 94% fewer tokens, 3.5x faster execution, 100% success rate. We break down how it works, why it uses the accessibility tree, and what the 'less is more' finding means for your website.
Cloudflare /crawl Endpoint: One API Call to Crawl Any Website
Cloudflare launched a /crawl endpoint that crawls entire websites with one API call — returning HTML, Markdown, or AI-extracted JSON. We break down what this means for AI agent readiness: why your robots.txt, sitemap, semantic HTML, and server-side rendering now matter more than ever.
Playwright: From Test Runner to AI Agent Interface
After Playwright overtook Cypress, Microsoft released Playwright MCP as an important browser automation option for compatible agents. We explain how getByRole and accessibility snapshots expose useful roles and names without guaranteeing complete agent navigation or task success.
AI Crawlers Ignore llms.txt — But AI Agents Don't
Dries Buytaert's data shows zero AI crawlers use llms.txt. But he measured the wrong thing. Crawlers scrape for training data — agents complete tasks. We break down why the crawler vs agent distinction matters, which coding agents already use llms.txt and content negotiation, and what you should implement today.
Anthropic's AI Exposure Index: What Real-World Usage Data Means for Your Website
Anthropic's new 'observed exposure' metric reveals a 61-point gap between theoretical AI capability and actual usage. We break down the data — from 75% task coverage for programmers to 14% hiring slowdowns for young workers — and explain why this adoption gap is a countdown for website AI agent readiness.
Build for Agents: Why CLIs Are the New Distribution Channel
Andrej Karpathy argues CLIs are exciting because they're 'legacy' technology — battle-tested, standardized, and universally parseable by AI agents. We break down why CLIs, MCP servers, and machine-readable docs are becoming the primary distribution channel for software, and what product teams need to build right now.
Explore more
Explore the checks behind measured stable-core coverage.