Every few days right now, someone I respect publishes a version of the same take: WordPress is finally dying. Cloudflare’s EmDash launch on April 1 lit the fuse. Mullenweg’s response the next day fed it. The supply chain attack on April 6 that backdoored 31 plugins kept it going. Then Joost de Valk, the founder of Yoast, said out loud that WordPress is “overkill for a lot of small businesses,” and the group chats lit up again.

I’ve been in WordPress my entire career. I built a plugin business on top of it. I’m still building one. I want to push back on the obituaries, not because the data is wrong, but because the data is being read wrong. WordPress isn’t dying. It’s disaggregating, and the part of it people keep pointing at as a death spiral is closer to a freeze. That’s a very different thing to plan against, and I think almost everyone quoting the numbers right now is quoting them with a conclusion glued to the front.

What the data actually says

Katie Keith at Barn2, who runs one of the most respected plugin companies in the ecosystem and who’s thought harder about this than most people, published Barn2’s 2025 survey in December. New plugin sales were down 17.8% year over year. That’s not overall plugin sales. That’s new sales. Existing customers renewed, which is why revenue was only up 0.65%. Barn2 is doing okay, and the Barn2 team have been pivoting into adjacent markets for years specifically because they saw the maturity curve coming. A founder I trust running a business I respect, weathering a storm she saw years ago.

A few months later, Blocksy cited a survey showing 48.8% of plugin companies saw sales worsen in 2025. Nearly half. That’s the number that’s been passed around the WordPress community for a month now, and it’s the one most likely to get quoted out of context.

Then W3Techs showed WordPress’s share of the web dipped to 42.5% in early 2026, the first meaningful decline in years.

Those three numbers, taken together, are what the “WordPress is dying” crowd points at. I don’t think they’re completely wrong. I do think, like everything else in this current era, it’s getting blown a bit out of proportion. The number is significant, but the conclusion people are drawing from it is the conclusion that a market is in terminal decline, and that isn’t what I’m seeing.

And to be honest, I think there’s a lot of discourse on the web from people who want WordPress to fail who are loudly declaring the demise of the platform just because they want that to be true.

My bet on what’s actually happening

I’m not a mind-reader, and this is mostly anecdotal. This is my bet. It’s the theory I’m operating on, built from watching the plugin market up close for a decade, running one of these businesses myself, and talking to a lot of other founders in the space. If I’m wrong, I’ll update. But this is how I’m framing the situation right now.

The 48.8% doesn’t hit every kind of plugin equally. It’s lopsided, and the lopsidedness is what the “dying” narrative misses.

Some plugins are getting absorbed. Some are getting commoditized out of the market. Some are getting displaced by SaaS. And some are doing fine, or even better than fine. Treating the average as a verdict on the whole platform is a reading mistake, and it’s the mistake every “WordPress is dying” post is making right now.

I do believe the “golden era” of building a WordPress plugin, getting free organic reach from WordPress dot org, and building a business on the back of WordPress using a series of well-written tutorials and candid blog posts is long dead. AI didn’t kill that though, it’s been like that for most of the 2020s. The market has matured, the ocean is red, all while AI systems are making most of the use-cases for a plugin evaporate.

What I think is actually going on, broken into the kinds of plugins I see hurting versus the kinds I see holding, goes like this.

Plugins I think are hurting

Start with the ones my gut says are getting crushed. Take a plugin that adds a contact form. Or a pricing table. Or a countdown timer, a testimonial slider, a cookie banner, a sticky header. One feature, implemented politely, shipped as a twenty-dollar-a-year plugin. These were fine businesses for a long time, because the buyer couldn’t write the feature themselves and didn’t want to. That condition is softening because the alternative got dramatically better. Claude will write you a working contact form block in an afternoon, tailored to your exact fields, your exact webhook, your exact compliance requirements. The plugin still works. The plugin isn’t a bad plugin. It’s just that the amount of complexity a user can absorb in a solution tailored to their exact scenario has gone up significantly.

Next, the AI-wrapper plugins. “AI blog post generator.” “AI meta description writer.” “AI image alt text.” These were never long-term defensible. They’re bridges to an underlying model, and the underlying model is now a browser tab away. Why would I pay thirty dollars a year for a WordPress skin over GPT when I’m already paying for the thing it’s skinning? The bridge is superfluous. I don’t expect these businesses to survive 2026.

And the simple utility plugins. The “replace one word with another” plugin. The “add this snippet to the header” plugin. The “redirect old URLs to new ones” plugin. Again, not because they’re bad. Because their job can be done by pasting a prompt into any coding assistant and hitting enter. That gap, between “I could write this myself if I knew how” and “I can write this myself now, because Claude does,” is the gap that’s closing.

Ironically enough, I even think that some managed hosting platforms are in for it, too. I loved WPEngine and Flywheel when they first came out, but their value proposition of an affordable, managed host that actually makes my life easier has been whittled away by ever-worsening support, increased pricing, and the fact that Claude can help me configure and set up a managed hosting setup that’s faster, cheaper, and easier to support. This one is definitely not for everyone but it’s another example of how AI is raising the threshold.

Plugins I think are fine

Now look at what isn’t hurting, and it tells you a lot about where the real line is.

WooCommerce looks fine to me, and so does Gravity Forms at the complicated end of their use cases, and so does Siren. These aren’t features. They’re substrates. You don’t replace them with a prompt, because the thing they do isn’t a feature, it’s a platform that other features sit on top of. Ask Claude to rebuild Gravity Forms and you’ll spend a month and still not have what you had.

The integrator plugins look fine too. The MailChimp-to-WooCommerce sync. The WooCommerce-to-QuickBooks bridge. The Stripe reconciliation utility. You can’t prompt your way around a well-designed integration, because the work isn’t in the interface. It’s in the edge cases, the error handling, the six years of nobody noticing when a refund cascades wrong. AI will happily write you the happy path. It will not write you the part that actually matters.

And the business-logic plugins feel safe. Membership engines with dunning and proration look fine. So do booking systems that handle timezone-aware cancellation rules, and LMS platforms with prerequisite gating. These encode decisions the buyer can’t articulate well enough to prompt in the first place.

The pattern I keep landing on is consistent. Products built as monoliths for 2015 WordPress are the ones I see hurting. Products built as portable logic that happens to deploy into WordPress are the ones holding. Even better: products built specifically so someone (or some AI agent) can extend them are the ones I think are going to win this era outright.

The plugins that win from here are the ones designed to be extended

I believe that the winners from here on out aren’t plugins that solve one problem. They’re plugins that reduce the friction of AI agents solving specific problems for specific buyers. Infrastructure wins. Platforms win. Plugins designed to be extended, by developers or by AI, win.

This isn’t abstract. Siren is the example I know intimately, because I built it. Siren isn’t one affiliate program with a commission rate. It’s a toolkit of well-designed primitives, with a robust REST and MCP surface explicitly designed to be hookable. A lot of the hard work when building Siren was making it genuinely extensible, because I wanted operators (and later, their AI agents) to be able to build programs I never thought of.

That work paid off earlier this month. A customer came through my support chat with a novel concept: they run a subscription site and wanted a checkout dropdown where the customer picks a registered school or club to receive a five-dollar donation, and they wanted to track those donations automatically through Siren. There is no off-the-shelf plugin for that. The closest thing would’ve been cobbling together three or four WooCommerce extensions plus a custom field plus a Zapier workflow, and even then it would’ve been duct tape.

Instead, they connected Beacon to Claude, Beacon pointed Claude at Siren’s engagement triggers documentation, and Claude generated a custom extension that registered a new engagement trigger, bound it to a WooCommerce checkout event, and mapped the customer’s selection to a Siren collaborator.

They dropped it onto their site the next day and it worked on the first try. A non-developer with an ex-developer background shipped a real, surgical Siren extension in an evening, without me writing a line of code for them.

That’s the shape of plugin that wins this era. Not a mountain of plugins each doing twenty things the buyer didn’t ask for. One extensible plugin, deep and well-factored, that an AI can learn and build on. The hard work of the plugin author was making Siren extendable, not making it feature-complete. The feature was the customer’s job, and they did it in an hour because the substrate was designed for exactly that.

I built PHPNomad specifically to hedge this bet

I’d been feeling this shift for a while before AI really took off. Back in 2021, I was already starting to watch the WordPress market mature in ways that made me uneasy. Big plugin companies were getting bought out. New product launches were slowing. Incumbents were consolidating. The market was becoming a place where you could keep going, but not easily start. I don’t know if I’d have articulated it this cleanly at the time, but the feeling was that WordPress is not a market I want my entire business living or dying on.

So I built a framework called PHPNomad specifically to hedge. Nomadic means the same core logic runs as a WordPress plugin, as a standalone microservice, as part of a SaaS, or as anything else PHP can host, without rewriting the product. Beacon, Siren’s MCP server, runs on Railway as a microservice right now. I could compile the same core logic into a ZIP and ship it as a WordPress plugin tomorrow, and it would work. That portability isn’t a feature I bolted on. It’s automatic if you follow PHPNomad’s conventions. You get it for free.

Siren has been nomadic from day one. I built PHPNomad specifically to build Siren, and I made that choice in 2022 because of the pattern I was already seeing in 2021. Any business should hedge, I think, and PHPNomad was one of the biggest pieces of hedging I was able to do. The bet has turned out to line up neatly with the AI shift, but it wasn’t primarily an AI bet. It was a “WordPress is getting mature and I want to own my own trajectory” bet.

My bet was made on a hunch that the monolithic WordPress model wouldn’t hold long-term, and that hunch is playing out right now, in public. When the 48.8% number landed, I didn’t panic, because I’d been building for this outcome for years without knowing exactly what shape it would take.

WordPress Ten Years From Now

I think that the internet is grossly overestimating how fast legacy WordPress sites will migrate. Big companies are too deeply invested in their WordPress stacks to even think about moving for the next decade. I’m working on some of those sites right now. They’re enormous, they make money, and they have absolutely no desire to make a big, monumental, expensive, and risky shift to a different platform yet. The reasons just aren’t strong enough to bother.

And of course, all of that assumes WordPress itself stops evolving, which I don’t actually think it will. WordPress has always been a set of extensible primitives and events, and AI has made it dramatically easier for developers and operators to bend those primitives into the thing they actually want. That’s a stronger case for WordPress in the AI era, not a weaker one.

The worst-case scenario is that WordPress freezes instead of adapting and ends up becoming a COBOL for the web. Even that isn’t a death certificate. COBOL didn’t die because mainframes died. COBOL froze. The mainframes kept running, the code kept working, and the market around them stopped growing new plugins because nobody new was showing up. That’s what the 17.8% looks like to me in its worst-case reading. Not a dying market. A potential freeze line. The existing installs will keep renewing, which is exactly why Katie’s renewal numbers at Barn2 still look healthy while new-license volume softens. New buyers aren’t walking into the plugin aisle the way they used to, because the new buyers are increasingly building what they need directly, or picking a tool that was never a WordPress plugin in the first place.

Legacy WordPress sites will still be running, and still making money, in 2036. The monolithic plugin era ending is not the same thing as WordPress ending.

This pattern is coming for every category AI touches

The reason any of this matters outside the plugin author crowd is that WordPress is just where the pattern showed up first, because WordPress had the most exposed surface. Every SaaS category. Every developer tool category. Every creative tool category. Every workflow tool category. The same sort of disaggregation is coming for all of them, because the thing that’s actually changing isn’t WordPress. It’s the cost of writing custom code, which just collapsed for anyone who can drive an AI agent well.

The question for every founder isn’t whether AI hits their category. It hits every category. The question is which bucket you’re in when it does. Are you building a feature people will soon be able to prompt into existence? Or are you building the substrate that makes their prompting possible?

If it’s the first one, start thinking hard about what your real defensibility is. If it’s the second, keep building, and make sure your extension surface is as clean and well-documented as you can make it. The winners of the next five years aren’t the plugins with the most features. They’re the plugins designed so someone else (developer or AI) can build the features on top.

WordPress isn’t dying. It’s splitting into parts, and most of the platform is going to keep running for another decade whether the obituary crowd likes it or not. Build for the parts that are going to matter. Leave the freeze line to the consultants.