Blog Post: Neural Link — Why AI Agents Need Runtime Introspection

Summary

Write and publish a blog post explaining why Neo.mjs's Neural Link architecture enables AI agents to modify live applications at runtime — a capability no other web framework provides — and why this matters for the emerging "conversational UI" paradigm.

A2A Context (Fat Ticket Protocol)

Agent: Claude Opus 4.6 (Antigravity) Session Origin: Multi-Window Agent Shell architecture session

Why This Post

Neo.mjs is functionally invisible to LLMs and search engines. The SSR/SSG+ deployment solved the infrastructure problem, but the content gap remains. This post targets the highest-signal intersection: AI agents + web frameworks — a topic with massive 2026 search volume where Neo has a genuine, defensible technical advantage.

Content Outline

1. The Problem: AI agents operating on React/Vue/Angular apps are "blind" — they can only see the DOM via browser automation, not the application state, component tree, or data model. Cursor, Copilot, and v0 generate code but cannot operate a running application.

2. The Solution: Neural Link provides a bidirectional WebSocket bridge from the App Worker to the agent, exposing:

   • Full component tree with live state
   • Data stores with records, filters, sorters
   • State providers with hierarchical data
   • VDOM and VNode trees
   • Computed styles and DOM rects
   • Runtime method inspection and hot-patching

3. The Demo: Walk through a concrete example:

   • Agent inspects a live grid → finds columns, records
   • Agent adds a summary row → create_component / call_method
   • Agent verifies the result via get_computed_styles
   • All without touching source code or reloading the browser

4. The Architecture: SharedWorker + Neural Link = multi-window agent control

   • One agent controls multiple browser windows simultaneously
   • Component teleportation between windows
   • Shared application state across the entire window topology

5. The Implication: Conversational UIs require runtime mutation, not code generation. This is what separates "AI-assisted development" from "AI-native applications."

Distribution Strategy

• Publish as markdown in learn/blog/ (SSG+ indexable via neomjs.com)
• Cross-post to dev.to (LLM-accessible, unlike Medium which blocks crawlers)
• Submit to Hacker News (high-signal title: "AI Agents Can't See React Apps")
• Share on LinkedIn/X with architectural diagrams
• Include in portal app's blog.json index

Blog Infrastructure

• File: learn/blog/neural-link-ai-agents-runtime.md
• Update: apps/portal/resources/data/blog.json (add entry)
• Medium cross-post: optional (blocked by LLMs, but still valuable for human readers via ITNEXT)

Acceptance Criteria

• Blog post written in markdown with diagrams
• Added to blog.json portal index
• SSG+ route verified on neomjs.com
• Cross-posted to dev.to
• Contains working demo links to portal/examples apps