Matt Mullenweg CEO of Automattic: What is The Future of WordPress in a World of AI?

Sure. My name is Matt Mullenweg, talking to you today from Houston, Texas where I was born and raised. I guess 23 years ago now, I connected with Mike Little over the internet on our blogs and we co-founded what we now know as WordPress. That little pebble that turned into a snowball that turned into an avalanche of code and community and plugins and themes and everything. I consider myself very, very lucky to be part of that. My other sort of night job is I do a little bit of investing with RG Capital, and my main gig is the CEO of Automattic, which makes WordPress.com, WooCommerce, Jetpack, Tumblr, Day One, Simple Note. We're a distributed company with about 1,400 people across the world, and yeah, I'm excited to be with you all today.

Well, in terms of threats, I guess the main thing I would think about is just that it's tied to the opportunity, which is that as we start to apply the coding part of AI, which has gotten quite good if you haven't been following with Claude Code and the new models like Codex, MiniMax, there's a lot of amazing stuff happening. So as that starts to be pointed at not just core but the 60, 70, 80,000 plugins and themes, I think we'll find lots of opportunities to improve. I think we're also going to find a lot of security bugs and things like that. So I would say like many open source projects around the world are being a little bit flooded with this sort of agentic driven input. And so the biggest thing we have to figure out over the next 18 months is how to incorporate this and deal with this both for core and our translations and everything we do in sort of the main WordPress and our official canonical plugins, but also how to provide the tools so that all the tens of thousands of the maintainers of these different open source plugins are able to sort of triage and incorporate all these improvements. Now at the end of this what we'll have is a vastly upgraded from a security, performance, and I think functionality as well, set of open source code and extensions which is very, very, very exciting. So yeah, I think like many things the opportunity and the threat is kind of like two sides of the same coin.

Well, it's nuanced, right? So, I do think that we're going to see a lot more people doing what WordPress is designed for, which is customizing it and making it bespoke for their needs. Which is I think beautiful. Now, essentially what you're doing though is you have some proprietary code that is unique to you. So you have to really balance the sort of gradient of whether it's something that's truly unique to your site and should be sort of proprietary code to your site or whether there are some sort of benefits to using shared code that other people are looking at updating, securing, maintaining, etc. So, I know that's a little bit abstract, but you think of it like, well, let's just use e-commerce as an example. I don't think you want to vibe code your e-commerce solution, right? Because you're going to have like compliance and credit cards and anti-fraud and all these sorts of different things. It's probably good to use some code that a bunch of smart humans have looked at and a company's backing and things like that. Now if it's just a different functionality, this is true in technology, the baseline of what people expect or what's possible is going to change. What's a good example? Let's just talk about image editing or all the stuff around managing images or libraries of images. AI is so good now at manipulating images, stuff you used to spend hours in Photoshop doing, you can now do with a single command. It takes like 10 seconds. So any plugins that used to exist around being better image editors or things like that, that's just something that's not as uniquely differentiated before, but it provides an opportunity for them to figure out what is uniquely differentiated. So this is going to happen a little faster than has happened in the past but it's not dissimilar to what is sort of the normal sort of gales of creative destruction as Friedrich Hayek would call it that is prevalent in being in the technology industry in general.

Well, I think you nailed the core thing that although the realm of what is possible is being vastly democratized, which I love. I mean that is literally the mission of WordPress is to democratize publishing. So these things being accessible to more people is possible, but also I mean you know it as you're following it like man it's hard to keep up with all the different AI models and what's launching every day. Sometimes they mess up and it can be very hard to debug sometimes and they can delete things. There's actually still someone described it almost like a slot machine, where sometimes it does something super incredible and you're like wow that would have taken me two months to code manually and it just happened instantly and then sometimes nothing happens or something bad happens. So the beauty of this is that for sort of motivated technical people, I think you get like not even just 10x, maybe 100x in terms of what you're able to do or maintain. And so agencies, I think their ability not just to serve more clients, but maybe even serve their existing clients much, much better than they have in the past, it's going to be increased. Now, if you're one of those clients, let's say you're a small business owner and your day-to-day and your things you want to think about every day might be serving your customers. That's a barber shop, a restaurant, a law firm, a doctor, whatever you might be out there, a teacher. You probably don't want to have to become an expert in using Claude and Codex and all these other things. You might have an expectation that when you call your agency, what used to take him a week now might take him a day. And I think that's completely fine. But there's still a benefit to having a phone you can call, someone you can trust to manage all this complexity, all the change and all the unknowns, sort of guide you through. Again, the AI can do anything, but you have to ask them the right questions. And so that ability to know what the question is and prompt them well I think is still something that for folks who really dedicate a lot of their time to it and make it their craft, profession, thing they love to do, there's going to be value in that.

That's a good question. And I'll say here my thinking here has probably shifted over the past maybe year and a half where I started talking maybe at least two WordCamps ago about remember I mentioned the Perplexity browser which is called Comet and more recently I talked about Atlas which is the OpenAI one. And so this idea that you have a really smart personal AI which is embedded into your operating system or your browser, I think just given the sort of context that's allowed, that is going to know you better and be more powerful. So it's also probably something you're paying for directly versus going into the hosting margin of your hosting plan or something like that which by the way it's not huge already. You're paying like dollars for something which is pretty amazing 24/7 globally available etc. So just to give a simple example, like when you log into WordPress.com, we'll know everything about your sites, and things you chose to connect or share with us, but we don't know all of your calendar. We don't know all of the chats you've had with your personal AI in the past. If you imagine this personal AI has access to all your photos, your emails, your everything, it's a very intimate relationship that you would be rightfully hesitant to ask for something. Honestly, with WordPress.com, I don't want to ask for all that stuff for you. So that personal agent could then essentially be the mediator. It can have access to all that super private stuff, manage it securely and have all that context and then talk to the tool, whether that's WordPress.com or any other website and just give it what that external site needs to have or needs to know. And actually, I'm really excited that in the future, some more of these models might be entirely local, as we get like new chips from Apple and others that can run actually some pretty powerful things just on your device on the edge, which of course from a privacy and security point of view is fantastic. So that personal agent I think will be able to do things on your behalf including using some of our existing interfaces. Like yes we're doing all the MCP and command line interfaces and all that sort of stuff and we actually have decades now of really strong API support everything but I think also these agents will be able to figure out any website even if they haven't built something specifically for AI they'll just start using it like a human will it'll just be slower. So in terms of what that balance will be I think it's probably going to be 80 or 90% it's going to be perhaps your personal agent using the tools. And so what it behooves us as tool makers, whether that's WordPress or others to do is make sure that as we design for human usability, which of course we're going to continue to improve, that we're also thinking about sort of the agentic usability. So how is this accessible to agents to use it and more like AI or agent driven things on behalf of humans using the tools and services.

Yeah I think it's actually awesome and it's a project that Peter and I have a lot of mutual friends and we've talked a bit through his whole process here. Also I think if you notice WordPress is one of the few accounts that OpenClaw follows. So that's no accident. I think there's a lot of parallels to OpenClaw and WordPress and I also think there's a lot of ways for OpenClaw to drive WordPress and I highly encourage you check out the wp-pinch project from Nick Hamsey. It's on GitHub. Which sort of connects your OpenClaw to your WordPress, allows you to do lots of management, different things. So, it's also got me thinking about, you know, OpenClaw loves working with command line interfaces. And of course, we have a long history of it's an official project. We brought it into the fold, I think eight or nine years ago, WP-CLI, which some of you are familiar with. So, I've definitely been thinking a lot about how to expand our CLI interface to expose some things that we might not currently support through it. That might be a way, this kind of old Unix philosophy of piping together lots of smart command line tools. Actually is a very, very efficient way for AI to interact with applications like WordPress. So, it's exciting to watch. We'll see where it goes. Things happen faster now than ever. Including adoption of things, switching between other things, alternatives. I've been following some of the there's things like NanoClaw or IronClaw or some things inspired by OpenClaw. I'm kind of rooting for OpenClaw. I kind of love the weirdness and quirkiness of the community and everything. And in fact, at Automattic, we're contributing some code to it and some security patches and other things. So, I think it's an interesting project to follow and we're certainly going to try to support it however we can.

Ah, that's a good question. Well, one, I would say that all the AI Frontier Labs, they're spicy, right? Like, there's definitely a lot going on behind the scenes that we're not aware of, but you can kind of see from the outside when people move between one or the other. They're also in a very unique position. I don't envy them like they're the targets of nation states and geopolitics and countless billions or trillions in funding and value that's being sort of going between them in a hyperly competitive environment. In terms of Peter going to OpenAI or any one of the frontier labs, I don't think it's a they are some of the most interesting companies in the world right now whether that's Anthropic or OpenAI or so and they have an access to resources which is a little bit unparalleled. So if there's something that if he wanted access to like the latest and greatest models and unlimited tokens and everything like that not to mention that there's just good sets of people at these things all my interactions with folks at these companies. They're definitely a magnet for a lot of interesting talent. So we'll see what happens. I agree that these companies sometimes have a mixed relationship with open source, but to be honest, you could say that of any large company. You could say that of Google. You could say that of Facebook. You could say that of Automattic even. We're probably one of the most open source forward companies in the world. Companies I think it's good we sort of anthropomorphize them and speak of them as a single entity but you have to remember these are often made up of thousands of individuals and they make mistakes those individuals make mistakes they change direction and they might do simultaneously things which think of Google and Apple they're like the biggest competitors in the world they also have tens of billions of dollars of partnerships and deals that they do every year in fact Apple now I think choosing the Gemini model to power some of the new AI stuff in their devices. So, I think as adults you have to look at this as like these companies are sometimes going to compete. They're also work together and it's good not to completely write them off or think that they're angels just at any given point in time like it's going to be complex just like humans and evolve.

Yeah. And sort of I like to call it decoupled rather than headless, but same idea. Using WordPress as a backend to talk to it via APIs is something we've supported very strongly. It's been a huge use case. Some of the largest WordPress sites in the world run this way. And I think that's the first kind of version, the v1. And I like to say we are in the command line DOS era of these AIs in a lot of ways was a lot of them creating sort of random stuff from scratch. I think the next generation is going to be a lot more saying like, oh, how do they build on existing libraries, existing open source projects in the most efficient way so they can use their tokens, use their customization to benefit from large existing communities and plug into existing tools. So I would say some of the most interesting stuff happening there as you might have noticed is we're starting to publish evaluations and benchmarks for WordPress. So basically this is a way for the frontier labs or tools like Replit, Lovable, v0 etc sort of say how well they work at creating WordPress blocks, WordPress plugins, WordPress sites etc. Right now there could be a lot of basic errors. I saw an example the other day where it created like a full plugin but it messed up something in the readme file so it actually wasn't loaded correctly. It was like a one-line change that was just like a very, very simple thing but it was a bug. So it made the hard part of the plugin it did like thousands of lines of PHP code and everything like that and it messed up something in the text file. So these are kind of basic errors as we provide sort of ways for AI to test their work and evaluate they will get better and better at sort of making no mistakes part of building things for WordPress and integrating with WordPress.

Yeah, there are WordPresses out there with hundreds of millions of rows in the wp_posts table. Yeah, with actually hundreds of millions of wp_users table like it can scale quite a bit do very, very complex things. The final thing I'll say on that eval is like one amazing thing and I've been pounding the drum about Playgrounds and for those who don't know basically Playground is a way that we spin up a fully containerized virtual machine inside your browser. It takes depending on how fast your computer is anywhere from 20 to 45 seconds with a database, web server, everything like that, PHP and it runs WordPress in this little container. It's completely sort of atomic so you can roll back forward so if you make a mistake you kind of roll back to a previous state in an atomic way and with checkpoints sort of along the way. So it's very hard to break if you break it. It's very easy to fix. Also because it's so cheap and easy to spin these up, it's free. Essentially, you run it in just the memory of your browser. And so, as the AIs use things like Playwright or other ways to sort of launch virtual browsers, they could actually spin up 20 of these, for example, and use it to verify whether the code they've written works. They can take screenshots and see whether it does the thing that it's expecting or show that to the user, say, is this what you wanted? So right now I mean we're doing some stuff with TX and others. the more like WordPress-centric AI things and I know some others in the community are but I think once we get the more frontier models and bigger tools to understand and be able to use things like Playground as a tool, it will unlock an incredible sort of amount of capability for this going forward and things that would have been unimaginable for like for example, let's just talk about plug-in compatibility. So, there's like 70,000 plugins out there and WordPress users have probably experienced where one plugin messes with another plugin, right? And right now, finding that out is a somewhat manual process. Like someone installs it, something breaks, they report it to the respective developers, they fix it. Now while we do do some automations a day for like the top thousand plugins or whatever and how they interact with each other obviously as that number goes larger the factorial combinations of every possible plugin being installed with every other plugin it's very, very difficult to do. You get into tens of millions or hundreds of millions of combinations. However, we could start to do that much more efficiently with a sort of Playground-like interface where we're actually spinning up different things and running certain automated tests in a much more efficient way than we currently do this with like virtual machines or other ways. Also we can do some sampling. So knowing sort of like some of the statistics of which plugins are most frequently used with others to sort of make this a little bit more of a O log problem which is like a little more efficient. So that I think is again a possibility when we think over the next 18-24 months like some of these things will start to be much better. So that's a lot of fancy words but basically a way to say that expect your WordPress in the future to not just be more secure but also just be a lot more stable as you install different plugins and try different things.

I've been thinking about this a ton. So, I'm glad you brought it up. So, first I'll say, yeah, there are going to be 100,000 plugins. We're just seeing the submissions go up. We already have a distinction in the plug-in directory which we introduced a few years ago I believe where there's sort of like a personal plugin so essentially solo developer there community plugin which means it has multiple contributors from a community and then there's commercial plugins right which typically have some sort of paid upgrade and it's typically a single company behind it. So I would love to see that second category the community plugins grow quite a bit. And right now it's a little easier if there's an existing plug-in and you want to do something slightly different for you to fork it or make your own version versus sort of collaborate on that more shared codebase. So, I've been thinking a lot about what are the collaboration tools we can give just like WordPress itself encourage people to maybe work together when they have a similar goal and purpose. So that we could have some consolidation perhaps of plugins which are again non-commercial but highly overlapping. Does that make sense?

Hey you make a good point. So the things I'll put in there that we're trying to balance is one I mean the biggest reason to use WordPress over those other proprietary platforms is that infinite flexibility and customization. The fact that the marketplace is so much bigger. So let's not forget that. Two, I would say that the thing we're trying to balance there is helping people find those more canonical or popular plugins like you said, so they're not presented with a bajillion options, but also making it possible for new things to come in, right? So that's the other thing you want to balance. Like you could easily design a system which points everyone towards the top old existing stuff. However, part of the beauty of an open marketplace is the ability for something new to come along and gain that traction, gain that user base, become the next Jetpack or Yoast or one of these things we think about as like the huge canonical plugins. So I think we definitely want to make sure the marketplace dynamics allow for incumbency but doesn't lock it in. So, but there's still a discoverability of newer things because we want to keep that open. And I've been doing a little bit of work here talking to some folks about ways we could actually perhaps use some editorial there to highlight some newer projects, ones that have really excellent code design, usability, etc. To sort of shine a little bit of a spotlight on some of these new things that people put in the work to make them excellent.

Oh man, this is a good question and probably actually one good for us to sort of wrap up on. So, we got a couple more questions. I might have to run. So I dropped something in the chat there. So one thing I'll say is that even if there were going to be no more sort of iterations with the new generations of chips and data centers coming online the capabilities of the models today when you talk about the diffusion of their impact on the economy is still relatively small compared to what I think could happen so that's basically another way to say that although these platforms might have like 800 million users or ChatGPT or something the way people that are using them is like seems not basic but they're using it to help with their homework and translate things and do things that aren't yet transforming companies or the economy. However, we're starting to see that sort of adoption by enterprises and humans adopting this thing is going to take a little bit of time. So, and that's a big topic right now is the diffusion of these impact on the economy. Actually a good discussion about this is Dario Amodei the CEO of Anthropic which was Claude which is the big coding one had a great interview with Dwarkesh Patel the other day which I think I blogged on ma.tt so if you check out my blog ma.tt blog about a lot of interesting AI stuff so it's a good place to stay current with what I think is sort of like the bleeding edge of this thinking. So I don't know if I agree as much with like a Gary Marcus or Yann LeCun who think that this technology isn't going to get us to where we're hoping there's a big, big impact. Perhaps for some definitions of AGI or ASI artificial super intelligence they might be correct but I do think that we sort of gradient descent or whatever will get us a lot even better functionality what we have today. And some of this is just again applying the existing functionality in different ways like a reasoning model or a model of experts or having AI check each other or think for longer or all these sorts of different techniques you could do to sort of get better results out of even some existing models or weights. In terms of the impact on the economy like is everyone going to lose their jobs etc etc. There has been predictions around that for almost every big technological revolution whether that's electrification or automobiles or all sorts of things railways like when you think of like the big society shaking changes refrigeration so yes some of those jobs go away actually there's a I saw a great YouTube the other day that talked about the whole way that ice used to be distributed and moved around the world and you had an ice box and there was an ice man who would come and deliver your ice every day and ice boxes used to literally be like you have a block of ice on a thing. So that was how refrigeration worked so before we sort of figured out how to deal with electricity and everything else. So yes the ice deliverer is probably a less big job now there still are some right you get the fancy ice at certain cocktail bars or something but it's not like something we're all getting delivered to our homes every day or every other day. So that sort of stuff is definitely going to change. However, think of the countless jobs and other things that were created. So I am typically a believer in like yes there will be technological disruption and certain jobs will change. However, there's going to be a lot of demand for new things and second that also the ability to learn or be retrained or reskilled is sort of it's the golden era for that. Like we know historically like the very best way to learn and be educated is sort of this idea of one-on-one tutoring and the infinitely patient again with today's models you can talk to them and learn anything you want. It is incredible. And so if you have a CSS bug, if you want to learn about physics, if you want to learn about whatever, I encourage you to using your AI as a tutor. And any of the models are actually really great at this today is kind of an incredible way to sort of increase your general edification and knowledge and be that lifelong learner, which I think is also I want to say like the skill like when I have friends who come to me with kids that are being born or going to college, they're like, 'What do we study? Like what do we do?' The thing I always come back to is learning how to learn and sort of yourself with a curiosity. Also I think a study of the classics philosophy logic more ethics those things like that I think that is going to become more important than ever. Certainly command of the English language and vocabulary I think is going to be more important than ever. But that sort of learning how to learn is a lifelong skill which I think is never going to go out and make you well adapted for whatever is going to come.

Yes and no. So one, bubbles aren't always a bad thing. So in fact if you look at things like railway or broadband in the 2000s the dotcom bubble like we think a lot about the pets.com and the internet companies but actually a really amazing thing was the amount of fiber that was being laid down that wasn't utilized right then but that dark fiber then became the basis for why things like YouTube, Facebook etc were able to grow and that dark fiber being lit up over the coming decades. So there was an overinvestment in the beginning but there was a huge consumer surplus from that. People like to say that there's not very many dead GPUs right now like basically the demand for this is exceeding our ability to produce it and even old generations are still having a lifetime and being used far longer than I think was originally predicted with a normal depreciation model of these assets. So that points to the demand for this perhaps being able to fill up the sort of idea of a Jevons paradox of like the demand filling up even what we seem like is a lot of supply. And then the final thing I'll say is there probably is some posturing or politics or fundraising where if you look at some of these crazy numbers being thrown around, some of it is announced and there's a competition to announce that you're going to invest more or perhaps you get like more kudos from a politician by saying you're going to make a $500 billion investment or something. I would not be surprised if along the way for the next few years we figure out some efficiency improvements and the way we're running some of these algorithms or inference or other things that perhaps means that we don't need to spend a trillion dollars of new things or 100 gigawatts of new power and we'll figure out some ways to get some similar growth and meet the demands perhaps in more efficient ways than the current sort of approaches and algorithms suggest. There's some very cool companies around there, not just inference companies like Cerebras, but like Liquid AI and others that I think are doing some really interesting work here.

Ma.tt is my blog. And actually, besides the newsletter or RSS, you can also subscribe. There's a Telegram channel, which is a really cool way to sort of get my updates as I post it. And also I think I post my photo blog is there as well. And then on like Tumblr, X, Instagram, everything on photomat. And actually a good place to follow me now is also GitHub. I'm doing a lot more stuff there. Been coding a lot more. And so my username there is the letter M.