---
title: "Why a 17-Year WordPress Agency Built Its Own CMS With Claude"
date: "2026-07-17"
video_id: "7663296482145029389"
tiktok_url: "https://www.tiktok.com/@tjrobertson52/video/7663296482145029389"
creator: "@tjrobertson52"
category: "Coding"
tags: ["wordpress", "headless cms", "claude", "emdash", "payload cms"]
excerpt: "TJ Robertson's agency stopped building WordPress sites and built a Next.js CMS with Claude in a weekend. What they tried first and what the claim leaves out."
---
TJ Robertson's agency has built WordPress websites for 17 years, and about half of its clients still run on it. In a TikTok from @tjrobertson52 he says, "I don't think we'll ever build another WordPress website again," because his team used Claude to build its own CMS in a weekend. The direction is credible, but the video shows none of the new CMS, so the weekend-build claim is untested.

## Why does he say WordPress is not AI native?

Robertson agrees that you can connect AI to WordPress and that MCP servers for WordPress exist. His point is that a connection is different from a platform an agent can run end to end. His test is practical. Claude can build you an HTML website directly, but it cannot build you a WordPress website the same way. When you connect AI to WordPress, he says, "it fights you every step of the way."

WordPress has AI plugins and integrations. Its architecture (PHP templates, a database-bound page builder layer and a plugin ecosystem) predates AI agents and was not designed for them to operate. An AI-native stack is one an agent can read, modify and deploy end to end.

## What alternatives did the agency try first?

The agency spent six months evaluating AI-native CMS options. Robertson names two:

- **EmDash**, the open-source, Astro-based CMS that Cloudflare launched in 2026 as a WordPress successor, with a built-in MCP server.
- **Payload**, the TypeScript-based headless CMS.

Neither one stuck. During the same period the agency's web dev lead, Hana, built her own CMS. Robertson cites that as a sign the team expected to leave WordPress before it had a replacement.

What changed the plan, he says, was Claude Fable, Anthropic's coding-focused model. "Fable is so good at software engineering, we've been able to build a ton of software." One of those projects became the agency's production CMS.

## What is in the custom CMS?

Robertson says Hana built the agency's ideal CMS in a weekend using Claude Fable. He lists these features:

- Every WordPress feature the team cared about, including a drag-and-drop page builder
- Pages built as HTML with Next.js rather than PHP templates
- Direct integration with each client's knowledge base on GitHub
- Easy migration of existing websites into the system
- Hosting on Vercel, set up so clients own their website while the agency keeps access

On lock-in, he argues that the output is plain HTML and JavaScript files, so moving to another CMS later would be easy. A small team of AI-native developers now maintains the system, and over the next six months he expects to recommend that many existing clients switch.

## Is a weekend-built CMS a real WordPress replacement?

Parts of the argument hold up. Cloudflare launching EmDash as an explicit WordPress successor shows a large infrastructure company betting the same way. Robertson also limits his own claim: he will not move e-commerce clients yet, which concedes that WooCommerce-level stores are hard to replace with a weekend build.

Other parts are unproven:

- **The weekend is the first build.** A CMS also needs auth, permissions, backups, security patching and editor training. WordPress spreads that cost across millions of users. One agency carries it alone.
- **Bus factor.** A CMS maintained by one agency is a risk for every client on it, even if the output files are portable.
- **No product on screen.** The video is a walking selfie with captions. It has no screen recording or demo, so there is no way to judge whether the drag-and-drop builder matches what WordPress clients are used to.

## Bottom line

Robertson's case is that AI coding now makes a purpose-built CMS cheaper to maintain than a 20-year-old platform that resists AI tooling. His agency evaluated EmDash and Payload, then built a Next.js CMS on Vercel with Claude Fable. The idea is plausible for brochure sites, he excludes e-commerce for now, and the video offers no footage of the CMS working.

## Sources

- [EmDash CMS](https://www.emdashcms.com/) - Cloudflare's open-source, Astro-based AI-native CMS
- [EmDash on GitHub](https://github.com/emdash-cms/emdash) - MIT-licensed source code
- [Cloudflare debuts EmDash (SiliconANGLE)](https://siliconangle.com/2026/04/02/cloudflare-debuts-emdash-challenge-aging-wordpress-ai-native-cms/) - launch coverage, April 2026
- [Payload](https://payloadcms.com/) - the TypeScript headless CMS the agency evaluated
- [Next.js](https://nextjs.org/) - the framework the custom CMS builds pages with
- [Vercel](https://vercel.com/) - where the client sites are hosted
- [Claude](https://claude.com/) - Anthropic's AI assistant, credited with the build


## Video description and verification notes

TJ Robertson has built WordPress websites for 17 years and about half of his agency's clients still run on it, but he opens this video by declaring the platform dead for his business: "I don't think we'll ever build another WordPress website again." His reasoning is that WordPress is not AI native, and after six months of evaluating alternatives, his team used Claude to build its own CMS in a weekend instead. **The core claim is that AI coding has gotten good enough that maintaining a custom, purpose-built CMS now beats renting features from a 20-year-old platform that fights AI at every step.**

## The Argument: WordPress Isn't AI Native

Robertson's central complaint is specific. He acknowledges that you can connect AI to WordPress and that MCP servers for WordPress exist, but argues that connection is not the same as being AI native. His test is practical: Claude can build you an HTML website directly, but it cannot just build you a WordPress website the same way. In his words, when you connect AI to WordPress, "it fights you every step of the way." The transcript renders the tool as "Cloud," a mishearing of Claude, Anthropic's AI assistant.

This framing matters because it separates two things that often get conflated. WordPress has AI plugins and integrations, but its architecture (PHP templates, a database-bound page builder layer, a plugin ecosystem) predates AI agents and was not designed for them to operate. An AI-native stack, by contrast, is one an agent can read, modify, and deploy end to end.

## What They Tried Before Building Their Own

The agency spent six months evaluating AI-native CMS options before committing. Robertson names two: EmDash (transcribed as "M-Dash"), the open-source, Astro-based CMS Cloudflare launched in 2026 as a WordPress successor with a built-in MCP server, and Payload, the TypeScript-based headless CMS. Neither stuck. Their web dev lead, Hana, also built her own CMS during this period, which Robertson cites as evidence the team already knew WordPress's days were numbered even before they had a replacement.

The real unlock, he says, was Claude Fable, Anthropic's coding-focused model. "Fable is so good at software engineering, we've been able to build a ton of software." One of those pieces of software became their production CMS.

## The Build: A Custom CMS in a Weekend

Robertson says Hana spun up their ideal CMS in a weekend using Claude Fable. The feature list he describes:

- Every feature they cared about from WordPress, including a drag-and-drop page builder
- Pages built as HTML with Next.js rather than PHP templates
- Direct integration with each client's knowledge base on GitHub
- Easy migration of existing websites into the system
- Hosting on Vercel, structured so clients still own their own website while the agency gets access

He also addresses lock-in directly: because the output is plain HTML and JavaScript files, migrating to another CMS later would be easy, though he doubts they will need to. The system is now maintained by what he calls a small team of AI-native developers, and over the next six months he expects to recommend many existing clients switch over.

## What the Visuals Show

This is a walking selfie video shot outdoors in a desert suburban neighborhood, with Robertson talking directly to camera in glasses and a patterned shirt. An orange title card at the top of the opening frames reads "I'm done with Wordpress - what we're switching to," and the only other graphics are word-by-word captions with orange highlight on key terms like "AI," "solutions," and "e-commerce." There are no screen recordings, demos, or shots of the actual CMS, so the entire pitch rests on his narration. Nothing in the video shows the weekend-built CMS working.

## Where the Argument Holds and Where It's Thin

The direction of the argument is credible. Cloudflare launching EmDash as an explicit WordPress successor shows major infrastructure players betting the same way, and "AI can now maintain custom software cheaply enough to beat platform rent" is a thesis showing up across the industry. The caveat Robertson adds is also telling: he is not yet willing to move e-commerce clients, which is an honest admission that WooCommerce-class complexity is still hard to replace with a weekend build.

The thin parts are the unshown claims. "Built in a weekend" describes the initial build, not the long tail of auth, permissions, backups, security patching, and editor training that CMS platforms amortize across millions of users. A custom CMS maintained by one agency is a bus-factor risk for the clients on it, even if the output files are portable. And the video shows no product, so there is no way to judge whether the drag-and-drop builder is comparable to what WordPress clients are used to. The claim is plausible but entirely unverified on screen.

## Key Takeaways

- A 17-year WordPress agency has stopped building new WordPress sites, arguing the platform is not AI native and fights AI tooling despite plugins and MCP servers existing.
- They evaluated EmDash (Cloudflare's Astro-based WordPress successor) and Payload before building their own CMS with Claude Fable in a weekend.
- The custom stack: Next.js pages with a drag-and-drop builder, GitHub-backed client knowledge bases, Vercel hosting, client-owned sites, and portable HTML/JS output to avoid lock-in.
- Robertson's implicit thesis is that AI coding flips the build-vs-buy calculus for agencies: maintaining purpose-built software is now cheaper than platform overhead.
- He excludes e-commerce clients for now, and the video shows no footage of the actual CMS, so the weekend-build claim goes untested on screen.

## Resources

- [EmDash CMS](https://www.emdashcms.com/) - Cloudflare's open-source, Astro-based AI-native CMS mentioned as "M-Dash" in the transcript
- [EmDash on GitHub](https://github.com/emdash-cms/emdash) - MIT-licensed source for the WordPress successor
- [Cloudflare debuts EmDash (SiliconANGLE)](https://siliconangle.com/2026/04/02/cloudflare-debuts-emdash-challenge-aging-wordpress-ai-native-cms/) - launch coverage, April 2026
- [Payload](https://payloadcms.com/) - TypeScript headless CMS the agency evaluated
- [Next.js](https://nextjs.org/) - framework their custom CMS builds pages with
- [Vercel](https://vercel.com/) - where the client sites are hosted
- [Claude](https://claude.com/) - Anthropic's AI assistant; the "Fable" model family credited with the build
