The Software Outlives the Economy
WordPress is entering its Drupal phase. What that means for the agencies built on it, and what the good ones do next.
Facts as of October 6, 2026. This expands “WordPress is entering its Drupal phase” on the Clapham Digital blog.
I. Everybody in my feed is selling something
Open LinkedIn on any given week and you’ll get one of two posts about WordPress. Either it’s dead, or it’s fine. Dead, with a chart. Fine, with a different chart and a little contempt for the people posting the first one.
Both camps are selling you something.
The “WordPress is dead” post usually has a product at the bottom of it. A new CMS, a site builder, an AI agency, a course on how to leave. The “WordPress is fine” post is usually written by someone whose market share, margins or headcount depend on it staying fine. Some of those people are making a reasonable bet with their eyes open. Some of them just don’t want to look.
So, in fairness, what am I selling? I make money whichever way this goes. I rescue broken WordPress sites. I move businesses off WordPress when it stops fitting them. I build new things on AI-first stacks, and I still build on WordPress where it’s the right tool. If WordPress came roaring back tomorrow I’d be fine, and if it disappeared I’d be fine. That’s why I can afford to tell you what I actually think.
I’ve also been inside it. I’ve worked in WordPress for twenty years, and until this summer I worked inside Automattic’s world, at Pressable. If you run a WordPress agency, I’m not writing this at you. I know what a care plan pays for, and I know how much of your team’s knowledge lives in WordPress-shaped habits. I’m writing it because I think both camps are handing you a bad map.
The two camps aren’t even arguing about the same number. “Fine” is looking at the installed base, every site already running, and the installed base really is enormous and really isn’t going anywhere fast. “Dead” is looking at new builds, where more and more of the people starting something today choose something else. Both numbers are real. The motive explains why each camp picked the number it did. It doesn’t tell you which claims hold up. For that you have to look.
II. The Drupal phase
Drupal was the serious CMS once. On W3Techs it ran 2.3% of all websites in 2018. Today it’s 0.6%. It’s still running government sites, universities, big publishers. People who make new things just stopped choosing it, and the market around it got thinner every year: fewer Drupal shops, fewer Drupal jobs, fewer people who’d recommend it for anything new.
That’s the phase I mean. A slow narrowing.
WordPress is at the start of the same phase, from a far bigger base. On W3Techs it sat around 43% of all websites from 2022 through 2025, peaking at 43.6%. Today it’s 40.1%. W3Techs’ sample shifted over that window (the share of sites with no detectable CMS rose), so treat that as a direction rather than an exact count. The HTTP Archive’s 2025 Web Almanac called it WordPress’s “first sustained slowdown after decades of expansion.” Commerce shows it more sharply. Store Leads counts just under 4 million live WooCommerce stores, down 10% year over year and well off the 4.78 million peak in mid-2024. Shopify is at 3.1 million and up 15%.
And WordPress still runs 58.6% of every site that uses a CMS at all. It’s going to hold most of that market for years.
So here’s the claim this whole piece rests on: the software outlives the economy.
Installs are sticky. A WordPress site that works today will very likely still be working years from now, because nobody migrates a working site for fun. The base will wear away at the edges, one rebuild and one rescue at a time, but slowly. What shrinks fast is the money. A huge share of sites isn’t a huge share of spend. The agencies, managed hosts, premium plugin shops, theme sellers and page-builder subscriptions all exist because the market was deep and growing, and I expect that economy to hollow out well before the installs do. The sites are still there. Fewer and fewer people get paid well to make new ones.
So what does that mean for a new build today? I think WordPress is no longer the best choice for most of them. There are real exceptions: commerce with custom needs, where WooCommerce’s flexibility still earns its keep, and domains where the WordPress ecosystem already solves the problem well and nothing else comes close yet. Outside those, every new WordPress build carries a risk worth putting in front of the client, which is that you’re building into a shrinking ecosystem and the bill for that comes later, to them.
III. Why WordPress can’t pivot
The natural question is why WordPress doesn’t just become AI-first. It has smart people, money, and more developers than anything else in the space.
It does have an AI story. There’s an Abilities API in core and an official MCP adapter alongside it, and people are doing real work there. The problem is underneath.
By AI-first I mean a stack an AI tool can read end to end: plain files, clear structure, one place where each thing lives, so a change can be checked against everything it touches. WordPress is twenty-plus years of nested backward compatibility. Hooks on hooks, plugins rewriting each other’s output, page-builder markup nobody wrote by hand. It can’t become that kind of stack without breaking the millions of sites built the old way. To get there it would have to break backward compatibility, or fork into a new product that doesn’t look much like WordPress. Both mean walking away from the installed base, and the installed base is the whole reason WordPress matters.
That’s the innovator’s dilemma, and the textbook version is about incumbents making rational choices for the customers they already have. Some of that is here. But I think there’s more to it than rational choice. Rebuilding would mean somebody deciding to break things for millions of sites they feel responsible for, and nobody in the community has been willing to make that bet. I don’t think that’s for lack of intelligence. I think it’s the will that’s missing, and the weight of the installed base is what ties it down.
We’ve watched this exact move before. Drupal 8 shipped in November 2015, rebuilt on Symfony components and moved from procedural code to an object-oriented architecture. It was the right technical call. It also meant Drupal 7 sites couldn’t simply upgrade, and a lot of them didn’t. Drupal 7’s end of life was set for November 2021, then pushed to 2022, then 2023, then to a final date of January 5, 2025, because so much of the base was still sitting on it. In 2013, two years before Drupal 8 shipped, developers who saw the rewrite coming forked Drupal 7 into Backdrop CMS, so those sites would have somewhere to go that wasn’t the rewrite. That’s what breaking compatibility costs a CMS with a big installed base. It isn’t hypothetical.
There’s a smaller illustration of that pull in WordPress itself. The block editor became the default in WordPress 5.0 in December 2018. Nearly eight years later, the Classic Editor plugin, which exists to switch it off, still shows 8+ million active installs. I’m not knocking the block editor. I’m pointing at how hard the installed base pulls back on any change, even one the project fully committed to.
The ecosystem used to be the moat
Whatever you needed, there was a plugin, and the plugin meant you didn’t have to write code. That was the value.
AI reprices it. When writing the code is cheap, “you don’t have to write code” is worth close to nothing, and what’s left are the costs: attack surface, update churn, plugins fighting each other, licenses that renew every year. A lot of what you’d install a plugin for, an AI-first stack does with a few lines you can actually read, or ships with out of the box. The ecosystem’s weight used to be a feature. Now you’re paying to carry it.
The best steelman I know for WordPress here: LLMs have seen more WordPress than any other CMS. They know the hooks, the APIs, the conventions. Surely that makes it the easiest platform to build on with AI.
It makes writing WordPress code easier. It doesn’t make a WordPress site legible. The state of a real WordPress site doesn’t live in files. It lives in the database, in the options table, in which forty plugins are active and what order their hooks fire in, in settings someone clicked two years ago. The model can write a perfectly good function. It can’t see the whole system, and mostly neither can you, so nobody can check the change against everything it touches. Knowing the language doesn’t help much when most of the program is hidden.
A second pressure, separate from AI
There’s another pressure on the WordPress ecosystem that has nothing to do with AI. I think it would be enough on its own.
Start with the record, which is public. In September 2024 Matt Mullenweg criticized WP Engine in his WordCamp US talk, and in a blog post the next day called it “a cancer to WordPress.” On September 25, WordPress.org blocked WP Engine from its servers. His stated case was that WP Engine gave back far less to WordPress than it took and traded on confusion with the WordPress name. WP Engine disputes that. In October 2024 WP Engine sued Automattic and Mullenweg, and 159 Automattic employees took a buyout offer. On October 12, WordPress.org forked WP Engine’s Advanced Custom Fields plugin into “Secure Custom Fields” under the existing plugin listing, citing its plugin guidelines and a security fix. In December 2024 a federal judge granted WP Engine a preliminary injunction. In January 2025 Automattic cut its sponsored contributions to the WordPress project to about 45 hours a week, saying it was reallocating resources because of the lawsuits. In April 2025 it laid off 16% of the company, and in May it announced its return to core. In September 2026 the antitrust claims against Automattic and Mullenweg survived a motion to dismiss, and the trial is set for October 19, 2027 by the court’s September 11 scheduling order. The same month, Automattic’s board voted to put Mullenweg on paid leave without a stated reason, and he was back as CEO about a day and a half later.
There’s a quieter part of the record too, and it’s the one I’d point a developer at. WordPress.org’s plugin search code is public. Text relevance gets multiplied by a set of boosts, and the heaviest is active install count. Plugins under a million installs take an extra penalty on top of that, and the code removes the overall cap on how much the boosts can add up to. There’s a fair argument for weighting by installs: it’s a hard signal to fake, and it keeps spam down. The side effect is that a new plugin that matches your search perfectly starts far behind the incumbents. Over the years I watched discovery for new plugins get close to impossible, and that code is the clearest explanation I’ve found for what I watched. (Jean Galea has a good breakdown of the maths.)
That’s the record. Now my read of it.
I think WordPress’s leadership has become a poor steward of its own ecosystem. To me, the public efforts to help discovery and lift up new developers look better on paper than they work in practice. And I think the fight with one of its biggest hosts, the ACF fork and the months of contributor hours pulled from core have split the community. From what I’ve seen, a real number of WordPress developers have left the platform over the leadership, not over AI, and some of them are openly rooting against WordPress.org now.
I don’t know anyone’s motives, and the case is still in court. What I can tell you is what the ecosystem looks like from where I sit: harder for a new developer to get a name into than it’s been in my twenty years, right when it most needs new people.
So I see two pressures, and they’re independent. AI reprices what the ecosystem was worth. The stewardship, I think, is driving off the people who’d have built whatever comes next. To my mind either one alone would hollow out the economy over time. WordPress has both.
IV. What’s actually being disrupted
This is where I think a lot of the “dead” crowd misses the point. They talk as if WordPress’s big gift was letting business owners control their own websites. That’s part of it. It isn’t how most sites actually get run.
Most business owners don’t have the time or interest to manage a website, and they shouldn’t have to. So they hand over what they want, their intent and their judgment about their own business, to an agency or a freelancer, and the agency builds it and keeps it running. Owner, then agency, then site. That chain is how most of the web works.
WordPress made that chain cheap and deep. An owner could find someone to take their intent and turn it into a site almost anywhere, at almost any price, because so many people knew WordPress. Plugins let the agency deliver quickly. The hosting market meant there was always somewhere to put it.
AI hits that chain from both ends at once.
V. The double squeeze
From one end, owners with AI tools leave. Not all of them, but more than you’d like. Someone who used to pay you to update their hours and swap a photo now asks a chatbot how and does it themselves. Someone who was going to pay for a new brochure site has an AI tool build one in an afternoon. And as they go, they stop paying what I’d call the platform tax, the managed hosting, premium plugins, theme license and page-builder subscription. Each cancellation is small. Together they’re the floor under a lot of agency business models.
From the other end, AI in your own hands collapses execution margin. Work you used to bill by the hour now takes a fraction of the hours. Keep billing hours and your revenue drops with them. Keep the old price for less work and someone down the street will undercut you.
It doesn’t land evenly. The bottom tier, the cheap sites and simple changes where there was never much judgment involved, is just gone. The middle gets pinched from both sides. The top holds, because the top is where the judgment is dense: work where somebody has to understand a business well enough to know what “right” means for it. If you want to come through this, climb toward that. A rough test of where you sit today: if a client could describe the job to a chatbot in a paragraph and get something they’d accept, that job is in the bottom tier, whatever you charge for it.
VI. The rescue wave, and why you shouldn’t count on it
There’s a hopeful story going around agency circles, and part of it is true. I predict a wave of rescue work.
A lot of people are going to take on their whole tech stack with AI and no qualified help. Some will do fine. Many won’t. They’ll ship sites with security holes, broken checkouts, data they shouldn’t be exposing, things that look great in the demo and fall over in month three. And the failure won’t show up where they’re looking. It shows up at the far end of the chain of trust. The customer hits it first, then the owner hears about it, and only then does anybody call an agency. I’m already getting some of those calls.
But I think a lot of that rescue won’t be fixing WordPress sites and handing them back. It’ll be moving them to AI-first stacks that give the owner more control and need less day-to-day help from anybody. That’s the right call for most of those clients, and it means every good rescue shrinks the demand after it. The client you move needs less care. And as better method spreads, fewer new sites break in the first place. The wave drains itself from both sides.
So it isn’t redemption, and it isn’t a return to the old margins and the old work. It’s a real but temporary bump from a first round of work done without proper method, and as that method becomes common, it dries up.
Take the work. Don’t build the business on it, and don’t hire for the wave. The agency that wins over the long run can take rescue work without needing it: lean, AI-first in its own workflows, with rescue as surplus rather than survival. And plan for your care-plan revenue to shrink, because the clients you rescue onto better stacks are going to need less care.
VII. The keeper of your client’s intent
So what’s left for an agency to sell?
“Judgment” is the easy answer, and it’s right as far as it goes. I use one word for the sharper version. Every job has a definition of “right.” Part of it can be written down and checked by a machine: the tests pass, the page loads, the form submits. Part of it can’t. It lives in somebody’s head: what this business is actually trying to do, what its customers will put up with, what good looks like for this particular site. I call that remainder the residue. When somebody holds the residue, AI makes them enormously more effective. When nobody holds it, the work still ships and looks fine, and the cost shows up later as slow decay nobody signed for.
Business owners carry a lot of residue and have very little interest in holding it in technical form. That’s what an agency is for now.
The way I’d put the job: you’re the keeper of your client’s intent. You understand the business well enough to hold their intent and judgment, translate it into the form AI works best with, and keep evolving the business with it. You turn what they want into action in the most cost-effective, token-efficient way you can, and you make sure the result matches what they wanted. You’re also the one who keeps the record of what was decided, why, and what “right” was supposed to mean. You hold the big picture for your clients and turn it into real outcomes, in a world where machines do most of the execution.
That’s a different promise from “we build websites,” and a more durable one. Durability is most of the point. The chain of trust runs from the customer to your client to you. Sell your client cheap work that quietly degrades and you aren’t only eroding their trust in you; you’re eroding their customers’ trust in them. A study of 6,299 repositories found 22.7% of the issues introduced by AI-authored commits still sitting there unfixed when the researchers looked. It has no human baseline, so it can’t tell you people do better. It does tell you how much was still sitting there.
Clients also need to be able to see the durability, or they can’t tell you apart from the shop that pipes everything through a model and ships whatever comes out. So show the receipts: written criteria for the job, independent review of what shipped, a record of the decisions, files the client owns, tests that run. That’s the difference between an agency and a pass-through.
AI can sell execution. Accountability is still yours to sell.
VIII. Plural tools, one discipline
If WordPress isn’t the default anymore, what is?
I don’t think there’s one default anymore. WordPress is a tool, and the agency that does well can use a range of them and pick the right one for the job: a static Astro site on Cloudflare for a brochure, Shopify for a simple store, WooCommerce where commerce gets custom, something else again for a membership platform. That kind of range used to be expensive, since every new stack meant months of a senior developer learning it. AI cut that cost way down. An agency that works well can now be competent on several stacks without a specialist for each.
It pulls against durability, though. Every stack you take on is another thing to keep healthy and another place for decay to hide.
My answer is to be plural at the tool layer and uniform at the governance layer. Use whatever fits the job, and run every job the same way: written criteria, independent review, a decision record and a clean handover, every time. Where you can, pick stacks that are AI-first in the sense above, so execution can run largely on its own. Your people’s time goes to settling the disagreements when the reviews don’t agree, and to holding the client’s intent.
IX. People and money
The hard question for agency owners: if machines do the execution, what happens to your people?
The work changes shape. Your developers stop being the hands and become the judges. The skill that matters is reading two reviews that disagree and knowing which one is right, and why, for this client. That’s harder than writing the code.
The real problem is the apprenticeship ladder. Juniors learned by doing the easy work, and the easy work is exactly what AI took. I don’t think you protect the easy work artificially. I think you reskill instead of deskilling: train juniors as reviewers and adjudicators from day one, close to the judgment, with seniors showing them why each call got made. My own trade, for what it’s worth: I’ve given up a lot of syntax I used to carry in my head, and in exchange I can run far more work at once and think more clearly about how systems fit together than I ever have. It’s like becoming the CEO of your mind instead of the sole operator of every step.
Pricing follows the same shift. Hours stop working, because the hours are disappearing. Reselling hosting and plugins stops working, because clients are cancelling them. What’s left to charge for is the holding: retainers for keeping the client’s intent, for oversight and accountability rather than labour. In short, an agency’s business moves from doing the work to vouching for it.
X. Where WordPress still fits, and the one bet I’d make
None of this means rip everything out. WordPress still fits a lot of real work. A store with custom needs. A site wired into a WordPress plugin nothing else replaces yet. A big content operation with years of editorial workflow built around it. For those, make it fast and stable, and plan.
Commerce is the interesting case. WooCommerce store counts are falling faster than WordPress overall, and I think that’s mostly simple stores leaving for Shopify, which is the right move for them. What’s left is the custom end, and that’s where money, inventory, fraud and trust all meet, and where AI-first platforms still don’t cover the ground well.
Agents are starting to buy things. OpenAI and Stripe launched an agentic commerce protocol in 2025. Google announced AP2. In January 2026 Google and Shopify launched the Universal Commerce Protocol with a long list of retailers and payment companies behind it. WooCommerce is in the game: it shipped an MCP integration and was a launch partner for Stripe’s Agentic Commerce Suite. But it’s a participant, not the standard.
There’s a version of the future where WordPress matters more, not less, and it runs through there. WooCommerce becomes the trust layer for agentic commerce: the governed, auditable core agents transact through, with WordPress rebuilt as an AI-friendly headless back end behind fast static front ends, and hosts as the managed layer for that governance. It would take exactly the kind of rebuild I described in section III. It’s a real opening.
I don’t see the will to build it yet. So I’m planning for the Drupal phase, and if someone proves me wrong, the agencies that already work this way will be the first ones ready for it.
XI. What I’d do this year
If you run an agency, this is where I’d start.
Reprice before the hours dry up. Move your best clients onto retainers for holding their intent, while you still have the relationship and the leverage. Do it by showing them the record you keep for them, not by announcing a new fee. If you wait until the hourly revenue has already gone, you’ll be negotiating from weakness.
Stand up one AI-first stack beside WordPress. Just one. Ship real client work on it this year. You don’t have to leave WordPress to stop depending on it.
Write down the criteria for every client: what “right” means for this business, in writing, before the work starts. That document is the residue made visible, and it’s what lets execution run without you watching every keystroke.
Put independent review on everything that ships. A second set of eyes, human or machine, that didn’t write the thing. Make it the rule and show the client the record.
Train your juniors as adjudicators. Bring them in at the judgment layer, because the typing layer is going away.
And don’t hire for the wave. Take the rescue work, and build for the market after it.
WordPress isn’t dead. It’ll be running a big share of the web for a long time. But the economy that grew up around it is hollowing out, and I don’t think the agencies that come through will be the ones who picked the right side of a LinkedIn argument. They’ll be the ones who worked out what clients were really paying for all along: somebody to keep their intent.
If you run an agency and you’re working through this, or you own a site and wonder which side of it you’re on, I’m at [email protected].