AI Didn’t Kill the Web, It Moved in! — Olivier Leplus (AWS) & Yohan Lasorsa (Microsoft)

Key Concepts

• AI Coding Agents: Autonomous tools that assist in software development by executing tasks, managing workflows, and interacting with repositories.
• Skills: Lightweight, text-based plugins that extend the capabilities of coding agents with domain-specific expertise.
• MCP (Model Context Protocol): An open standard that allows AI agents to connect to external tools, data sources, and development environments.
• Web AI APIs: Browser-native APIs (e.g., Summarizer, Proofreader, Prompt API) that allow developers to run AI models locally on the client machine without external cloud calls.
• Agentic Web: The evolution of web applications designed to be navigated and operated by AI agents rather than just humans.
• LLM.txt: A standardized text file (similar to robots.txt) that provides AI agents with a map and context of a website’s content.

1. AI-Enhanced Development Workflow

The presenters, Yohan and Olivier, emphasize that AI is now integrated into every stage of the web development lifecycle.

• Coding Agents & Skills: Modern agents use "Skills" defined in .agent/skills folders. These skills use an open specification to provide agents with specific capabilities, such as interacting with the GitHub CLI, recording videos via Playwright, or creating local tunnels (e.g., Telegram notifications) for mobile testing.
• Agentic Workflows: Developers can define an agents.md file to automate repeatable tasks, such as recording a video of a new feature, deploying a tunnel, and notifying the developer via messaging apps, ensuring the agent doesn't close a task until the developer confirms completion.

2. Debugging and Performance Tuning with MCP

The session demonstrates how Chrome DevTools MCP allows AI agents to control the browser directly.

• Capabilities: Agents can perform actions like clicking elements, filling forms, capturing console logs, and running Lighthouse audits.
• Performance Analysis: By controlling network throttling (e.g., 3G vs. Fast internet), agents can generate performance reports, identify LCP (Largest Contentful Paint) bottlenecks, and suggest optimizations like image resizing or CSS preloading.
• AI in DevTools: Chrome now includes native AI assistance in the console and network tabs to explain errors (e.g., CORS issues) and suggest fixes. Developers can even apply CSS changes made in the browser directly back to their source code via the "Apply to Workspace" feature.

3. Local AI APIs in the Browser

The presenters highlight the shift toward running AI models directly in the browser, eliminating the need for cloud-based API tokens and latency.

• Summarizer API: Allows developers to summarize text (e.g., product reviews) using different formats like "key points" or "TLDR."
• Proofreader API: Automatically corrects spelling and grammar in input fields, providing the corrected text along with metadata (start/end indices) for UI feedback.
• Prompt API (Multimodal): A general-purpose API that can process text, images, and audio. The presenters demonstrated an image-to-review generator that outputs structured JSON, allowing the AI to "see" a damaged product and write a corresponding review automatically.

4. Preparing for the "Agentic Web"

As AI agents become more prevalent, websites must be optimized for machine consumption.

• LLM.txt & LLM-full.txt: These files act as a roadmap for AI. LLM.txt provides links to documentation, while LLM-full.txt provides the entire content of a site or library, ensuring agents have the most up-to-date information, bypassing the "training data cutoff" issue.
• Web MCP: This experimental framework allows developers to register specific functions (e.g., add-to-cart or submit-review) as tools that an AI agent can call directly. By adding simple attributes to HTML elements, developers can transform standard forms into "tools" that agents can fill, validate, and submit without human intervention.

5. Notable Quotes

• "It's no longer the question of can I code my web app with AI, but rather how to get the best results out of AI coding agents." — Yohan
• "I tend to think that this is the same [as responsive design]. Make sure your website is going to be prepared once we have all the agentic browsers coming out on the market." — Olivier

Synthesis and Conclusion

The transition to an "agentic" web era requires a fundamental shift in how developers build applications. By leveraging MCP for tool integration, local browser AI for performance and content processing, and LLM.txt for discoverability, developers can create more robust, AI-friendly applications. While these technologies are currently experimental, the presenters argue that early adoption is essential to remain competitive, much like the shift to mobile-responsive design in the past. The ultimate goal is to create a seamless ecosystem where AI agents and humans collaborate efficiently within the same web environment.