ethos
← Back to Projects
🌐

Smart CMS

AI-powered website migration and generation

Thesis

Website migration is painful and expensive. AI can understand semantic structure (not just scrape HTML) and rebuild sites that clients can actually edit.

The Problem

Website migration and creation is broken for small businesses: Agencies charge $5-15k to rebuild a site that already exists somewhere else. DIY tools like Squarespace and Wix produce cookie-cutter results. Custom builds are expensive and leave clients dependent on developers for every text change. The real problem? Existing tools either scrape HTML (garbage in, garbage out) or start from blank templates. Nobody's teaching AI to understand what a nav, hero section, testimonial block, or pricing table actually is.

Implementation Approaches

Approach 1

Migration-First MVP (Recommended)

Recommended

Start with crawling and rebuilding existing sites, add generation later

⏱️2-3 weeks
📊High complexity

Implementation

  • AI crawler extracts semantic structure, not raw HTML
  • Outputs component JSON model (nav, hero, testimonials, etc.)
  • Auto-generates scoped visual editor for that specific site
  • Next.js build pipeline deployed to Vercel
  • One MCP integration (scheduling) to prove the hub concept

Pros

  • +Immediate value, clients have a site to migrate
  • +Real pilot client ready (pilates studio)
  • +Every crawl feeds the generation model later
  • +Proves the hard part (semantic understanding) first

Cons

  • High complexity, lots of edge cases in real sites
  • Need to decide how opinionated to be about fixing messy designs
  • Migration is one-time revenue, need subscription upsell
Approach 2

Generation-First

Build vertical-specific site generators, add migration later

⏱️2 weeks
📊Medium complexity

Implementation

  • Start with one vertical (fitness/wellness)
  • Curated component library based on best practices
  • AI generates sites from business description + preferences
  • Same editing layer and build pipeline
  • Migration comes later as crawl data accumulates

Pros

  • +Cleaner output, no messy source sites to deal with
  • +Subscription model from day one
  • +Can target specific verticals with tailored components
  • +Faster to MVP

Cons

  • Another website builder in a crowded market
  • No data flywheel until migration is added
  • Harder to differentiate without the migration angle
Approach 3

MCP Hub First

Build the integration layer, let sites come later

⏱️1-2 weeks
📊Medium complexity

Implementation

  • Standardized MCP servers for scheduling, payments, CRM, reviews
  • Works with any frontend initially
  • Position as integration infrastructure for the incubator
  • Sites become the delivery mechanism for integrations

Pros

  • +Integrations are reusable across all incubator tools
  • +Lower risk, integrations are well-defined
  • +Guildry and other tools benefit immediately
  • +Can validate integration demand without full CMS

Cons

  • Doesn't prove the core thesis (semantic understanding)
  • Integration-only business is harder to monetize
  • Need a frontend eventually anyway

Validation Plan

Hypothesis to Test

Small businesses will pay $99/mo for a site they can actually edit, migrated from their existing mess in under a week

Validation Phases

1

Pilot Migration

2-3 weeks
  • Crawl the pilates studio client's existing site
  • Build the semantic extraction pipeline
  • Generate Next.js site with editing layer
  • Deploy and get client feedback on editing experience
Client can edit headlines, images, and schedule without developer help
2

Expand to 3 Sites

2 weeks
  • Find 2 more pilot clients in fitness/wellness vertical
  • Run same pipeline, track time and edge cases
  • Document patterns: what components repeat, what breaks
  • Add scheduling MCP integration to all three
Migration time under 1 week per site, clients actively using editor
3

Pricing Validation

2 weeks
  • Propose pricing to pilot clients: $500 migration + $99/mo hosting
  • Gauge reaction, adjust based on feedback
  • Create landing page for waitlist
  • Run small ad test targeting local businesses with ugly sites
2/3 pilots convert to paid, 20+ waitlist signups

Kill Criteria

Stop and move on if any of these become true:

  • Semantic extraction quality is too inconsistent across real sites
  • Clients don't actually use the editing layer
  • Migration takes more than 2 weeks per site
  • No willingness to pay at target price point