I Vibe-Coded A Site On Lovable. Then I Moved It Somewhere I Can Actually Use It| Practical AI Ep 40

Practical AI: Episode 40

I Vibe-Coded A Site On Lovable. Then I Moved It Somewhere I Can Actually Use It.

Watch on YouTube

Published: May 8, 2026

TL;DR

  • Lovable has $400M ARR, 8 million users, and no native CMS. Their migration path to WordPress is duct tape. Olga built a sales site in Lovable, moved it to PageMotor live on the show, then edited it three ways without re-prompting any chatbot.
  • Chris shipped PageMotor’s full admin API at 6:15am the morning of the show. Authenticated admin access. Everything possible in the software is now possible via API. Point Claude Code at your site and tell it what to do.
  • Chris’s prediction of the year: “Everyone who cheats the great refactoring is going to get absolutely steamrolled.” Companies that lock into one AI model lose to platform-agnostic builders.
  • Pentagon picked 8 AI vendors and excluded Anthropic over Mythos cyber capabilities. Same week OpenAI launched The Deployment Company, a $10B PE joint venture.
  • Funding tracker: $10.1B for AI this week. 66% of total funding. Top 3 rounds = The Deployment Company $4B + Anthropic Applied AI JV $1.5B + Sierra $950M = $6.45B for installing AI INTO enterprises, not building models.

Table of Contents

About This Show

Practical AI is a weekly live show (Fridays 11am CT) hosted by Olga Pechnenko and Chris Pearson that cuts through AI hype to deliver news, trends, and hands-on tips for builders and founders. Unlike technical AI podcasts, Practical AI focuses on business applications and ROI—what actually works, what’s hype, and what you can implement Monday morning.

What You’ll Gain

  • Understand the deployment lane in AI. Three of the top 4 AI rounds this week aren’t building new models. They’re installing them into Fortune 500 workflows. See why $6.45B of weekly capital chose the installation business.
  • Learn what’s broken about the Lovable-to-WordPress migration path. Lovable has 8 million users and $400M in ARR but no native CMS. The “fix” is duct tape: third-party plugins, DOM-scrapers, manual React-to-PHP refactors. See the real failure mode in 5 minutes.
  • Discover what an AI-native CMS actually is. Not a chatbot pasted onto a 25-year-old codebase. A site you direct instead of edit, where AI agents do the operating work, where the website becomes a business partner instead of a file cabinet.
  • See a live migration in 8 minutes. A Lovable-built sales site, converted to PageMotor through the design conversion skill, then edited three times — price change, font change, header change — without re-prompting any chatbot.
  • Gain a strategic frame for picking your AI stack. Chris’s “great refactoring” prediction: companies that lock into one AI model get absolutely steamrolled by platform-agnostic builders. The winning move is near-zero switching cost.

Biggest Takeaway to Implement: If you’ve built something on Lovable (or Bolt, or v0) and you’re not sure what to do with it next, the answer is not “find a better WordPress importer.” The answer is choose your next destination based on what you actually need to do with the site. For most real businesses, that’s a CMS — and the AI-native ones are the only kind that won’t be obsolete in 18 months. Pick the destination first. Pick the migration tool second. 

PageMotor and Practical AI Updates

Free, informative, and FUN!

Frequently Asked Questions

Can you migrate a Lovable site to WordPress natively?

No. Lovable doesn’t have a native WordPress export. The migration path is third-party — tools like WPConvert.ai (ZIP-to-theme), CloneWebX (DOM scraping), or Elementor importers. Results vary; static marketing sites convert cleaner than apps with backend logic. Read more below.

What is “the great refactoring”?

Chris’s term for the structural rebuild every company needs to do to become AI-native and platform-agnostic. His prediction: “Everyone who cheats the great refactoring is going to get absolutely steamrolled.” Companies that lock into single AI models will lose to builders that stay model-agnostic with near-zero switching costs. Read more below.

What did Anthropic announce this week?

A lot. Colossus 1 compute deal with SpaceX (the underused data center after Colossus 2 came online), $200B Google Cloud commitment over 5 years, doubled rate limits for Pro/Max/Enterprise, and a $1.5B financial-services joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs. Read more below.

Why did the Pentagon exclude Anthropic?

Mythos. The Pentagon’s 8-vendor selection (SpaceX, OpenAI, Google, Microsoft, Nvidia, AWS, Oracle, Reflection) cited Anthropic’s Mythos cyber capabilities as a “supply chain risk.” Anthropic’s safety-first posture got them Wall Street and lost them defense in the same week. Read more below.

What is PageMotor’s new admin API?

Chris shipped it the morning of the show, May 8, 2026 at 6:15am CT. Authenticated admin access. Everything possible in the software is now possible via API. You can point Claude Code at your PageMotor site and tell it what to do — create landing pages, delete users, audit SEO, add structured data, run sales reports. Read more below.

What is an AI-native CMS, exactly?

A CMS where the AI is the operator, not a feature. The website becomes something you direct instead of edit. Olga’s frame: “The traditional CMS is a file cabinet. The AI-native CMS is a business partner.” Four shifts define it: the site adapts to your business in real time, adapts to each visitor, becomes a memory layer, and coordinates your stack. Read more below.

Practical AI: I Vibe-Coded A Site On Lovable. Then I Moved It Somewhere I Can Actually Use It.

Key Definitions

What is “inlining” HTML?

Taking everything that lives in separate files — HTML, CSS, JavaScript — and pulling it all into one single self-contained file. For React-based sites like those built on Lovable, this is the first step to migrating to any non-React platform, because the raw HTML source from a React app shows nothing visible (it’s rendered in the browser, not on the server). The skill PageMotor uses on conversion expects an inlined HTML file as input.

What is The Great Refactoring?

Chris’s term for the structural rebuild every company needs to do to become AI-native and platform-agnostic. Companies that “cheat” the great refactoring — by locking into one AI vendor’s tools and building workflows around them — will be outcompeted by builders that stay model-agnostic with near-zero switching cost between AI providers. Coined on air during Episode 40.

What is an AI-native CMS?

A content management system where the AI is the operator, not a feature. The traditional CMS (WordPress, Drupal, even Webflow) is a file cabinet you click through. The AI-native CMS is a business partner you direct. Four defining shifts: the site adapts to the business in real time, adapts to each visitor, becomes a memory layer that answers operator-level questions, and coordinates the entire stack from one interface.

What is “digital welfare”?

Chris’s term for the dependency mountain that modern React-based sites accumulate. A typical React app pulls in 200-300 third-party JavaScript packages, each requiring updates, each capable of breaking the site, each a security surface. The site is technically working but is being kept alive by a sprawl of external code the operator never approved.

What is a “loadbearing class”?

An invisible CSS class in the HTML that controls layout or behavior. In a highly designed page, deleting a loadbearing class breaks the page silently. PageMotor’s editor colorizes these — invisible/structural elements get a color tag so editors can’t accidentally delete them.

Quotable Moments

“It sounds like a lot of duct tape to me.”

— Olga on Lovable’s React + Vercel + Supabase + GitHub deployment stack

“It’s no longer about who has the best model. It’s about which companies you are the most ingrained in.”

— Olga on Anthropic + OpenAI’s simultaneous Wall Street JV announcements

“Everyone who cheats the great refactoring is going to get absolutely steamrolled. That’s my prediction about that.”

— Chris on companies locking into single AI vendors

“The website of the future is directed, not edited. A traditional CMS is a file cabinet. An AI-native CMS is a business partner.“

— Olga on the structural shift in how websites work

“Once the API is fully complete, you’ll have a conversation with your site and it’s going to change things the way you talk to it.”

— Chris announcing the PageMotor admin API at the end of the show

0:56 Lovable’s WordPress Bridge Is Mostly Duct Tape

Lovable made news this week by surfacing a path for users to migrate their Lovable sites to WordPress. Chris flagged it immediately as architecturally suspect. A Lovable site is a React app deployed through GitHub to Vercel, with Supabase as the database. Four different services and a build pipeline keeping each site alive. To migrate to WordPress, you need to inline all the rendered HTML, then translate the React components to WordPress templates, then re-implement any backend logic — all while inheriting a fresh stack of WordPress plugin dependencies.

Key Takeaway: “It sounds like a lot of duct tape to me.” — Olga, in the cold open. The duct tape image stuck and reappeared throughout the deep dive.

10:52 AI Beats ER Doctors at Harvard

A Harvard Medical School + Beth Israel Deaconess Medical Center study tested OpenAI’s o1-preview on real emergency-room cases plus clinical vignettes. The model scored 67% on diagnoses and management. ER physicians scored 50-55% on the same tests. AI beat doctors by 15 percentage points on the highest-trust use case in healthcare.

The Harvard Study

AI: 67% accuracy. Doctors: 50-55%. OpenAI o1-preview vs ER physicians on real cases and clinical vignettes. Published in the New England Journal of Medicine, May 2026.

Chris reframed the result through decision fatigue. Doctors make hundreds of decisions per shift. AI doesn’t get tired, doesn’t have shifts, and doesn’t have to remember every protocol from memory. The framing isn’t “AI replaces doctors.” The framing is “AI removes the decision-fatigue tax from clinical work.”

14:23 Anthropic’s Compute Scramble + $1.5B Wall Street JV

Anthropic owned the news cycle this week. Three independent announcements:

Anthropic’s Three Compute Stacks

SpaceX Colossus 1 (rented after xAI moved to Colossus 2, ~11% occupied), $200 billion Google Cloud commitment over 5 years, and the existing ~$100 billion AWS commitment. Three concurrent compute partnerships across three different infrastructure stacks.

At Code w/ Claude SF, Anthropic shipped Code Review, Claude Design (visual outputs), Memory for Managed Agents (public beta), 10 financial-services agent templates, and doubled Claude Code rate limits with peak-hour reductions removed. Then they announced a $1.5B joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs to embed applied AI engineers directly inside banks.

Pro tip: Olga’s frame for the week: “It’s no longer about who has the best model. It’s about which companies you are the most ingrained in.” The model wars are over. The distribution wars have started.

21:44 OpenAI’s $10B Deployment Company + Pentagon Excludes Anthropic

OpenAI mirrored Anthropic’s Wall Street play at larger scale. $10 billion joint venture with TPG, Brookfield, Advent, and Bain Capital — legally incorporated as “The Deployment Company.” OpenAI majority control. $4B raised so far. The JV sells deployment services to enterprises.

The Pentagon’s 8 Vendors

SpaceX, OpenAI, Google, Microsoft, Nvidia, AWS, Oracle, Reflection. Anthropic was excluded over “Mythos cyber capabilities as supply chain risk.” The same safety-first posture that got Anthropic a Wall Street JV lost them defense in the same week.

36:15 Cheating The Great Refactoring

The sharpest Chris monologue of 2026 so far. The context: with Anthropic, OpenAI, Google, xAI, and others now all pursuing enterprise lock-in via JVs and deployment partnerships, companies are about to pick “their” AI vendor. Chris’s prediction: anyone who locks in loses.

Chris’s prediction (~37:35): “Everyone who cheats the great refactoring is going to get absolutely steamrolled.” Picking one AI model and building your workflows around it feels safe today. It is the opposite of safe. The companies that stay platform-agnostic with near-zero switching cost between AI providers will outcompete the locked-in ones within 12-24 months.

The mechanism is straightforward. AI model quality is converging at the top. Pricing is converging. Capability is converging. The companies that win will be the ones that route work to whichever model is best at each task — and that requires architectural flexibility, not vendor commitment.

41:44 Lovable Has $400M ARR And No CMS

The deep dive opens with the rallying frame. Lovable hit $400 million in annual recurring revenue with 8 million users and 146 employees — confirmed by TechCrunch in March 2026. They built the fastest-growing AI website builder on earth. And they don’t have a native CMS.

The Lovable Gap

8 million users. $400M ARR. 146 employees. No native CMS. When their users want to manage the sites they built, Lovable routes them to WordPress, Contentful, or Sanity. The biggest AI website builder on earth uses legacy WordPress as its upgrade path.

Olga’s experiment: built two sales sites in Lovable to see what the migration problem actually is. First prompt: a one-line brief. Result: weak. Second prompt: a 13-point detailed brief (worked out with Claude). Result: better but still no business questions asked, random stock photo included, the page had a clear “made by AI” feel. Her verdict on air: “I’m sticking with Claude for any kind of vibe coding or not.”

52:20 The Live Conversion: PageMotor Skill On A Lovable Site

Chris ran the PageMotor design conversion skill against the inlined HTML export from Olga’s level-2 Lovable site at the top of the show. The skill ran in the background through news and funding. The reveal at 52:20 was honest: the CSS didn’t compile. An RGBA color value in the Lovable export hit an infinite loop in PageMotor’s SCSS compiler (which expected hex with alpha channel). Chris debugged live.

Pro tip from the debug: the conversion skill logs every failure mode to a failures.md file so the same error never breaks the same way twice. The first time anything fails, the system learns. By the time you’re the third or fourth user hitting an edge case, the skill handles it cleanly.

The honest read: live demos with visual reveals are fragile when the converted artifact has any error. But the debug itself made the moment real — Anastasiya watching from home now knows that PageMotor handles errors gracefully and learns from them.

59:34 Funding: $10.1B Week, Deployment Lane Dominates

AI Funding Week 23 (May 1-7, 2026)

$10.10 billion across 49 AI companies. 66% of all weekly funding. Cumulative through 23 weeks: $289.8B. Average weekly: $12.7B.

The top 3 AI rounds tell one story:

  • The Deployment Company — $4B Private Equity, San Francisco. The OpenAI / TPG / Brookfield / Advent / Bain joint venture in Crunchbase form. Direct news-funding crossover.
  • Moonshot AI — $2B Venture, Beijing, China. The Kimi people. Largest single-company China AI round of the entire 23-week tracker. China dollar share rebounded from 1.5% (Week 22) to 21.4% (Week 23) on this round alone.
  • Anthropic Applied AI JV — $1.5B Private Equity. The Blackstone / Hellman & Friedman / Goldman Sachs financial-services play.

$4B + $1.5B + $950M (Sierra’s Series E) = $6.45B for installing existing AI into Fortune 500 workflows. 64% of all AI dollars this week went to deployment, not model-building. The most expensive line item in AI is no longer the model. It’s the salesforce that gets the model into a bank.

1:19:51 Three Live Edits On The Converted Site

Chris demonstrated three real edits on the PageMotor version of the migrated site. Pricing card updated from $25 to $100 (clicked into the price element, typed, saved). Primary font changed to IBM Plex Sans and IBM Plex Serif (selected the heading, picked the font, saved). White header background added via a custom CSS rule (Chris’s lane — the live edit option for designer users).

Olga’s reaction: “Look how amazing that is. You don’t have to prompt anything. You go and change it and boom, you have a different site.” On Lovable, the same three changes would require three chatbot prompts, three regenerations, three deploys, and three opportunities for the chatbot to break something else on the page. On PageMotor: three clicks.

1:26:51 The Four Shifts: From File Cabinet To Business Partner

Olga revealed a slide she made the morning of the show. The strongest single positioning artifact PageMotor has produced.

The Frame: The traditional CMS is a file cabinet. The AI-native CMS is a business partner. Same design surface. Different category of product.

Shift 1 — The site adapts to your business in real time. It rewrites itself based on what you’re launching this week. Creates new pages when it notices repeated buyer questions. Changes homepage emphasis when business priorities shift. Flags outdated claims, pricing, bio screenshots, and stats — and fixes them.

Shift 2 — The site adapts to each visitor. Builds personalized paths through the site based on who landed. Creates micro-sites for different audiences without starting from scratch. For known prospects, the hero swaps to address the stated objection.

Shift 3 — The site becomes a memory layer. Answers questions about your business like an operator, not a chatbot. Explains your funnel gaps, content gaps, trust gaps, conversion leaks. Generates campaigns from what already exists inside the business.

Shift 4 — The site coordinates your stack. Reads from your CRM, email system, offers analytics as one system. Updates pricing the moment Stripe updates. Builds the entire launch funnel from a brief.

Chris’s synthesis: “A living presentation. Your living brochure. A living system that updates as your business does.”

1:32:27 PageMotor API Shipped This Morning

The “and one more thing” reveal at the end of the show. Chris announced that the full PageMotor admin API shipped at 6:15am the morning of the show. Authenticated. Everything possible in the software is now possible via API.

What this means in plain English: point Claude Code at your PageMotor site. Give it admin authentication. Then tell it what to do. “Make me a new landing page.” “Audit my SEO and tell me what to fix.” “What were last month’s sales of Product X?” “Delete the user with the username annoyingguy1992.” “Add structured data to these pages.”

This is what an AI-native CMS actually is. Not a chatbot pasted onto a 25-year-old WordPress codebase. A real CMS where the AI is the operator, not a feature. Phase 2 of the PageMotor bridge thesis shipped ahead of schedule.

Vibe coding gives you a pretty website. PageMotor gives you a real one — that an AI can run on your behalf.

Keep Learning