The CMS era is sunsetting.
Websites are becoming software again: built with AI, readable by AI, transacting with AI.
I have built the web since 1994, before HTML was a standard. I watched the CMS era begin, and I can read the data that says it is ending. This page is the argument, and the site you are reading it on is the evidence.
I · The numbers turned
The decline is on the record.
43.6%
WordPress share of all websites at its mid-2025 peak (W3Techs)
41.5%
Share by July 2026, the first sustained decline since tracking began in 2011
7,966
New WordPress vulnerabilities reported in 2024, up 34% year over year
96%
Of those vulnerabilities that came from plugins, not core
For fifteen years the story of the web was CMS adoption. WordPress alone grew to 43.6% of every website on earth by mid-2025. Then, for the first time since W3Techs started measuring in 2011, the line bent down: 42.6% by March 2026, roughly 41.5% by July.
A one-point drop sounds small until you remember nothing had ever made that line go down. Not Squarespace, not Wix, not Webflow. Those moved share between content management systems. This is share leaving the category.
The pressures are documented. A governance crisis at the center of the WordPress world: an active lawsuit, roughly a quarter of the combined workforce gone, the largest investor marking its stake down 63.5%. A security load that grows with the plugin economy itself: 7,966 new vulnerabilities in 2024, 96% of them from plugins. And one new pressure that did not exist before, which is the actual subject of this essay.
II · Why the CMS existed at all
One problem, and a permanent tax to solve it.
The CMS solved exactly one problem: programmers were a bottleneck. If a marketing manager could not change a headline without filing a ticket, the website stagnated. So we put a database and an admin panel between people and their websites, and non-programmers could publish.
The price was everything else. Themes that all look alike. Plugins stacked on plugins, each one somebody else's code running in your house. Upgrade treadmills, compatibility breakage, and a security surface so large that an entire industry exists just to patch it. Companies stopped owning their websites and started renting them from their own tech stack.
We accepted the tax because the alternative was worse. That arithmetic held for twenty years. It does not hold anymore.
AI did not make the CMS easier. It made the CMS unnecessary.
The bottleneck the CMS existed to route around is gone. A disciplined human directing AI can change production code as fast as a marketing manager can type into an admin panel, with version control, review, and a full audit trail that no CMS ever offered. When the constraint disappears, so does the machinery built to work around it.
The public conversation has not caught up. Most of the commentary frames the WordPress decline as "WordPress versus Webflow," one CMS losing to another. That is reading the scoreboard one level too low. The real move is CMS to code plus AI. Websites are becoming software again.
III · What replaces it
Three properties of a post-CMS website.
The order matters. How a site is made changes first, because the build bottleneck is what AI removed. How a site is read changes next, because ChatGPT, Claude, Perplexity, Gemini, and Grok are already the front door for a growing share of every category. How a site transacts changes last and matters most: the next visitor to your website is increasingly another machine, and almost no website can answer it.
IV · This site is the demo
Do not take the argument on faith. Test it.
No CMS anywhere
Written in code with AI under my direction, reviewed by me, shipped from git to production. Zero plugins, zero templates, zero admin panel.
A front door for machines
A discoverable agent card at /.well-known/agent-card.json and a live A2A endpoint. Point your own agent at it and ask about me.
A grounded agent for humans
The chat agent on every page answers from one knowledge base, the same source of truth the site itself is built from.
Why I wrote this
Every transition needs people who have crossed one before.
I have shipped through every era of the web since 1994: before HTML was a standard, the first live music stream, enterprise, federal, and now production AI. I am open to AI leadership roles, full time or fractional, where this transition is the job.