[00:00:00] Speaker A: Welcome to Within WordPress, your podcast. About WordPress with me today is James LePage. And James LePage, welcome. First of all.
[00:00:16] Speaker B: Thank you for having me.
[00:00:18] Speaker A: Great to have you here.
The. The timing of the recording of this podcast couldn't be more perfect because yesterday there was a huge announcement and I'd love for you to introduce yourself and then touch the announcement because I think, yeah, like I said, I think the timing could have not been any better.
[00:00:42] Speaker B: Yeah. And it's funny that it worked out like this and that announcement's timing was also interesting as well. So I'll introduce myself. I'm James lepage. I am currently the AI lead at automattiq and now also one of the co team leads and Reps for the AI
[email protected] before that, before all of this, I ran a startup in WordPress called WPAI for a few years. Before that I did another startup. Before that I did a WordPress agency and essentially kind of grew up in the WordPress industry and got my start as a freelancer and then grew into that agency. So I've been involved in WordPress for a very long time and it's very fun right now for me, taking my education in AI and my background in building companies and bringing it to WordPress the project and also Automatic, the company.
[00:01:39] Speaker A: Yeah. So the announcement was the introduction of the, the, the make team, the AI team, which we'll, which we'll cover later.
Can you share a little bit more about. Because you're the company that the startup that you created was acquired by Automattic, if I recall correctly. Right?
[00:01:57] Speaker B: Yes. Yeah.
[00:01:59] Speaker A: And can you, can you tell a little bit about what that role became inside of Automattic?
[00:02:05] Speaker B: Yeah. So first, the startup was a startup focused only on the WordPress industry and it was founded to build AI for this massive industry that I knew needed AI tools because I had operated here for years and years and years. I'd also studied AI in university. So that crossover of, hey, I know how to build some of this stuff and I know that I've needed this as an agency owner in terms of troubleshooting and support. And code generation led to the formation of what was originally a side project called CodeWP, which was a code snippet generator for AI. And it was released around when ChatGPT was released and that just snowballed into something that became very clear. There was a market need and demand enough to go and create a full company out of it. And I had done that previously outside of the WordPress industry with another startup as the technical co founder. So this time around went and raised money and grew this company for WordPress. And the main thing here was I'm from WordPress, I know that WordPress is this huge industry, I know it powers so much of the Internet, but many people don't get that.
So I never expected there.
And actually a part of the thesis of that startup was there wasn't going to be much in terms of AI for WordPress coming from traditional Silicon Valley. It would definitely come from people like me who came from the industry and other founders and agency owners within, at least in the beginning. So that was what we did. We did AI and WordPress for about two years, starting with code, but growing into more traditional consumer focused agents. And we started with code before cursor existed, started with some of the conversational stuff like one week before ChatGPT. So we weren't super ahead of that one.
And we started with agents before agents became this whole calling card of this is the new way everything will work. So we were a bit ahead in the context of WordPress and had always kind of operated in parallel with Automattic and a lot of the other larger players in the industry. And at one point it became really clear that we could be very impactful as people and bring some of our technology over to Automatic, which resulted in that acquisition and brought over our team and our products, rolled it into some of the work that was already going on in the company, namely the AI site builder that you shouted out in the beginning. That was a project that I kind of walked into and helped offer some guidance of like here's what we did, here's how the stuff we built worked and shaped some of that. But can't take credit for the project or the success of that launch.
But then also kind of looked at what was going on with AI at Automattic and said we could be doing a lot more, we could be building for many more surfaces in many different ways in WordPress. And over the past six months of me being at Automattic, I've kind of scaled what I'm focused on and what I'm doing into a overarching lead of the AI projects and focuses at Automattic and how we are building for the ecosystem of WordPress, WordPress.com, wooCommerce, but also for the other products as well. Parsley Day 1. All of the products that Automattic houses.
And that's really what I've been focused on at Automattic for the past six months.
[00:05:37] Speaker A: That's kind of what I expected, but it's nice to hear the whole totality.
The project that was released was the WordPress.com site builder, which I've played with. And from what I can gather, and I haven't, this is maybe 15 minutes I've played around with it, but it's quite solid. It's like.
I'm hesitant to call it really good, mainly because I see also limitations, but as a, as a first implementation, I was pleasantly surprised at how well it was functioning and how well it was actually listening at the things that you do. How much of a challenge was it?
Because I don't know if there's a word for it, but testing for negative prompts, meaning people just going to RAM and just stuff anything into that prompt, hit Enter, and then that's going to magically work. But you also have to exclude the things it shouldn't do.
How much? How difficult was that part?
[00:06:42] Speaker B: Very. Because when you build with AI, it's this almost unlimited input and users have come to expect I'm going to put something in and something is going to happen, happen. And so they'll put whatever they want into these products and it's up to the creators to make sure that there are good, accurate and safe outputs of whatever these, these tools are. So it's unique in the fact that we haven't really ever built software that works or interacts with customers in this way.
So you're almost designing the new interaction and the new UI that users will soon come to expect with everything and doing it in almost an experimental manner. And then when you pair that with the deployment on WordPress.com and the scale that comes with that and the eyeballs, it becomes a very difficult project. And there's a very incredible team behind that work, making sure that it is solid. And I think that the rating of solid when you look at some other AI products out there from other companies is a really good rating because these are very difficult products and if it does what is advertised, that's a huge win.
[00:07:51] Speaker A: Yep.
[00:07:52] Speaker B: Somebody. Somebody said and. And I think Big sky or the AI site builder, the internal name is Big sky, the AI site builder is very impressive already, but it's also the worst it will ever be. Like it's going to continue to improve, which is awesome.
[00:08:08] Speaker A: Exactly.
Many, many people hating on AI, on the quality of the. The thing they thought they requested, I should say, and I the emphasis on thought because prompting is indeed an art.
But whatever result they get, and I have friends and family who play with AI because most of them, certainly my kids use it in many different variations, but none of them understand the prompting as well. So they're kind of shooting from the hip and sometimes they hit and sometimes they miss, but they start complaining about the quality of the output and the thing that it produced and more importantly, the lack of the thing they expected to be also there.
And I tell them, look, this is. It's a prompting game. It's not.
So I actually don't like the word that we're calling it AI, because there's nothing artificial about it and there's no intelligence. It's all very much programmed.
[00:09:09] Speaker B: I think, also drawing on the prompting stuff. Yes, it is a prompting game, but it's also not the user's responsibility to study about how to prompt and how these models work and all of this. So I think, to a degree, I.
[00:09:24] Speaker A: Think, think there's you, you need to understand prompting better. But that goes in the same vein of as just using Google Pre AI.
[00:09:31] Speaker B: Yes, as a consumer.
[00:09:33] Speaker A: And they expect it to work. Exactly. No, you need to do a little bit more than the bare minimum.
[00:09:39] Speaker B: Yeah. As a consumer, I think the best thing you can do is learn how to prompt and learn how to even like removing yourself from AI. How do I logically present what I want? Like, how do I think about what I want and just present it logically? For the builders of AI, it's up to us to say, well, we can expect every single consumer, especially the market that the AI site builders are currently targeting, which are these basic sites, basic requirements. People who have no other alternative other than either trying to build it themselves or using this support assistant or assistant to build their sites. You have to kind of expect them not to know how to prompt. So how do you design systems in a way where you can walk them through the journey? Or how do you interpret whatever the users implementing to try to get to a good solution or go back and forth with the user to help shape their requirements. So I think it is almost a balancing act of, yes, the consumer should know how to prompt because this is the way you're going to interact with software into the future. But you also have to be a mindful creator of these products, especially in the transitory period of people don't know how to prompt and they're going to just like dump massive paragraphs and copy paste of brand books and all these different things. And it's up to you to say, do it this way or don't do it this way or let me process this or things like that.
[00:10:59] Speaker A: Yeah, I Agree.
So in a previous career I was a project lead, project manager for project that in some way produced a software application doing a particular thing in the field of hr.
And one of the things we quickly learned, and I'm telling this because I see a parallel in how you described how to build for AI, because I'm realizing it's essentially the same thing as with any piece of software because you have to build it in a fashion where you have to take into account the familiarity of the end user. So we built stuff for hr, but they were very much IT related.
They needed it, they needed a tool to work with it.
And essentially the thing is, we came to conclude that when we think it's ready as developers, the next step is to find somebody who has absolutely zero clue about what you built and you know, have them have their way with it. And that essentially results into, let's call it making it monkey proof. Like if I'm touching three buttons at the same time, does it still do what it's supposed to do instead of just hitting enter? If I hit shift enter, because I'm used to doing that, does it break things? And my word, they were really good at finding those types of weirdness. And I imagine AI and implementing AI solutions has only made that problem.
I don't know what the, what the fold number is, but 10 times worse, maybe 100.
[00:12:45] Speaker B: I would say. Yeah, it's just because you have so many surfaces and such a blank canvas for a user to interact with these tools with the conversational stuff. But I agree with you. The best product is a familiar and simple product that does what you want it to do. So if you can create that with AI, marrying it with existing software alongside new design, then you, then you have a really good experience that can scale up and down to the most simple request to the most complex request. And that's what is exciting about AI. It can serve multiple demographics of customer if you create it in the right way.
[00:13:23] Speaker A: Yeah, I agree.
If the. So I'm going to touch on one particular thing which I think you're the perfect candidate to explain to people who don't still don't fully understand what the power of an agent is.
Because I'm assuming there's a whole bunch of agents. Maybe it's one huge one or maybe there's different ones. I don't know how you architecturally design such a solution, but can you explain what an agent is and what the benefit of working with an agent is in the larger realm of AI?
[00:13:59] Speaker B: Yes, and it's funny because when we were first building AI, I hated the term agent because it didn't really exist as it exists today. And it slowly scaled and scaled and scaled until it really could become something that is an agent. But at the same time, it was kind of born a marketing term and never defined exactly.
So everybody has their own little definitions of what an agent is. For me, an agent is a system that has the agency to fulfill, discover, well, discover and then fulfill a user request. The goal state of the user and the way these things get put together varies on who's building them. But the underlying logic here is that this is a system that allows a LLM to have agency and determine, hey, the user has asked this, I have access to these specific tools, I have access to this data. What do I need to do to achieve this goal?
And that's probably the simplest way I can put it. You can have this overarching orchestrator where there are multiple sub agents that specialize in individual tasks. But at the end of the day, this is an AI, an LLM that has access to these tools and the resources and uses these tools and resources in a non deterministic way to achieve the user's goal state. So if I go and say, hey agent, go and make me a page.
The agent may say, well, I have the ability to go and write Gutenberg block markup. I have the ability to go and query information about that user from their WordPress site. I have the ability to ask the user a follow up question if I'm not confident in what the information they've put in is. And you could think about it in that simple three tool context. There's typically a lot more complexity there. But the system will go and say, all right, I'll go and render the block. I'll go and, and make sure that the actual markup itself properly can be injected into the post editor or the site editor. If it fails, I'll go and take that error and then go and think about the error and say why is the error happening? And then go and revise and kind of have the agency to work within the confines of the system to get to the goal. And that can scale out to.
It works for the page builders, but it also can scale out to really any system, any industry, any goal state.
[00:16:25] Speaker A: Yeah, I tried to explain. That's a wonderful explanation by the way.
I always try to add the context and capabilities context to what is an agent? Meaning there's context of what it is supposed to do and what it's not supposed to do and what sources it uses and all those things and then the capabilities of what it's allowed to be doing to sort of help define that there's sort of a circle of things you build it to be really good at and that potentially means it's not good at other stuff. Whereas a generic I log into ChatGPT or Cloud or whatever and I just ask it and it kind of figures out what I'm after.
And you know, there's not really a predefined agent.
Cursor will probably be a better version of an agent, but also that has different sub agents, I would say.
[00:17:17] Speaker B: Yeah. And I think it's also interesting because as the definitions of agents change based on who you speak with and the solutions that are being built and the way the industry is going, you could absolutely define claude, for example, paired with MCP tools as an agent and you could provide CLAUDE guidelines and these tools and these resources and say go and do something in this specific way. And it's probably not going to be as accurate or robust as this purpose built solution embedded into existing software. But the concept remains true where we are converging as an industry.
And now I can confidently say that these types of agent experiences do exist. We're converging on an industry, on standards like MCP and just means and methods of doing things that are these boutique implementations.
They're consistent ways and patterns that you use LLMs and you prompt LLMs to orchestrate sub agents or tools or collections of tools and resources in context which haven't existed until really right now. And it's fun because this is the way the industry works.
You consistently discover new ways to do things and converge on the solutions that work as an industry.
[00:18:28] Speaker A: Well, that combined with continuous upgrading of operator modes and whatnot.
You know, there's a huge difference between.
I'm mainly using ChatGPT, but I play with others as well. But like upgrading the same question for O4 to O3, that's quite a difference that you get in response. It's actually telling you how it's thinking. It explains how it's doing its thinking, its assumptions and all that and. And then it starts producing. That's wildly different from when ChatGPT was introduced.
[00:19:04] Speaker B: Like yes, yeah. And what's also very interesting is that all of this stuff is typically based in research that's published openly. So this is open research. That for example, the way these reasoning models are working, this research was occurring years ago and slowly became more and more formed and papers built on papers, research built on research, and eventually it ended up in these experimental models within the proprietary labs and the open source models and all converged on, hey, like this works really well and specifically with the reasoning stuff has become basically the standard model that are used, that is used within ChatGPT and these agents and these individual solutions. So it's fun when you come from kind of that background of understanding how the technology works and you can track it through these papers and kind of read the tea leaves to see where things may go in the future.
And also a lot of parallels with the open source world in that there is open source AI, but there's also a lot of open research and it's all available to you.
[00:20:11] Speaker A: Yeah, I like that you use tea leaves here as the, as a means to figure out what it is you're looking at.
[00:20:19] Speaker B: I like tea.
[00:20:19] Speaker A: Yeah, so do I. But you are aware of the free interpretation sometimes I can have on anything.
And that's, I think it's a funny parallel to actual tea. Leave reading one question on a definition which I'm aware what it is, but I'd love for you to explain what is an mcp?
[00:20:44] Speaker B: MCP is an open protocol and it's actually, it's a protocol, it's a set of published standards and it's a way for people to build what many consumers define MCPs to be.
This is a protocol that allows both software creators and AI clients to define, okay, I want to speak with software. So software will expose its functionality in this unified way. And I will look at this software and these MCP tools and resources and as an AI tool, I will consume this because I know it's unified and essentially to an LLM that is connected with software through these unified tool interfaces and resources.
So I find MCP as a builder very interesting because this is the first really widespread standard and protocol in AI land that allows all of these providers, in terms of the folks producing the end clients like Claude, like cursor, like ChatGPT, to go and interact with almost this decentralized collection of tool servers, which are things like GitHub, WordPress itself, exposing functionality, things that can be done on the software, data that can be read from the software in a unified way to really supercharge the consumer experience.
So that's kind of a backwards description of what MCP is, but I like biasing towards the standard first as opposed to what it unlocks.
[00:22:22] Speaker A: Could you, could you call an MCP the API for AI?
[00:22:28] Speaker B: Their definition is the USB C port for AI. And I like that terminology because it's this universal way you plug in functionality into this brain.
I will.
[00:22:42] Speaker A: USB C is built on APIs as well, so I, but I get it.
[00:22:46] Speaker B: Well, I'll try to push away from APIs because there's a big discussion and debate specifically within MC of like, okay, we're kind of just exposing at this point a lot of APIs that already exist in this unified endpoint, this USB C style plug. And there's a lot of discussion on like should we just be interacting with APIs? Like there's a fast API implementation for connecting to AI. So there's a lot of discussion behind a lot of this. And what interests me about the MCP and the discussion that surrounds MCP and other standards is this is how kind of industries have formed in the past. This is how standards have come about in the past. And AI is going through its own formative stage. So I love mcp. I think it's a very good standard and it's a very unique collection of the hive mind of AI forming solutions that need to exist.
But it's also very interesting to see the discourse around MCP and how these standards are being formed in real time directly in front of us.
[00:23:52] Speaker A: So yes, the directly in front of us, considering the short term, it's actually been fully available is wild because generally the standardization phase of any type of software solution is not within, let's call it 24 months.
[00:24:14] Speaker B: Yes. And it's really, in terms of it actually becoming useful has been three months and it's still not very useful because there are unknowns around the authentication, there are unknowns around the security implications of exposing functionality to software in this non deterministic LLM centric way. So it's very interesting this has happened massively quickly and it's clear that there are needs and demands from the industry, which is why this has happened so fast. But it's also fun to kind of watch the discourse of hey, how are we going to handle authorization and authentication?
And it's all there on GitHub discussions and within these products and these companies. And it's very fun to watch.
[00:24:58] Speaker A: Yeah, I get that certainly from your role. So I'm a consumer of AI and I build modestly stuff with it certainly for scanfully the tool that Barry and I are building we see a huge potential of what an MCP type of solution over the layer of data that we have to our availability, what type of wonderful stuff we can have AI do.
But it's, you know what, what I'm seeing and I'm, I'm guessing this is most people working with WordPress in one way or another still have a very limited view of what it actually is, where the future is going, what are the next steps, what are the next pivotal moments we see coming? I have less view on that. Could you, could you share a little bit about how that works in your world? Because you're, you know, you're, you're more than knees deep, you're, I'm guessing at some, at some points gasping for air.
[00:26:05] Speaker B: Yeah, I think that's a big question. It's, it's a question of like, where does AI go? And I think the most direct answer is that nobody really knows, including the folks at the proprietary labs like OpenAI and Anthropic. They're just going where the customer research and discovery and the models and the research that goes into unlocking additional capabilities brings them. So ChatGPT was built in a weekend and it was an experiment that grew into this massive accessibility boom of generative AI. And Anthropic started kind of like as the writing centric, the CLAUDE models were these writing centric models where you could use them to write in a more humanistic, human centered way. And Anthropic has very rapidly pivoted towards code and we're going to be the best code and developer focused lab here.
So I think all of that to say these are these billion dollar companies that are at the forefront of AI and they are pivoting like they are not billion dollar companies.
So all that to say like we are all discovering where this technology will go together and following what the models unlock and understanding what users actually want from this technology and research and use cases to go and build the most valuable things in terms of what consumers can expect. I think that we're seeing traditional software get married with LLMs and you see that through MCP connecting to these additional or these existing clients. You also see that with tools like Cursor becoming incredibly popular with developers and non developers alike, you see that kind of bubbling down into the bolt lovable style interfaces where now you can kind of paint with code, but you're not actually looking at the code, you're looking at more of I'm interacting with this AI tool and I'm getting the software out of it. You're seeing it with the mcp, but you're also seeing it with these AI integrations built into existing ecosystems in a native way. So there's a lot of fragmentation here because everybody's still experimenting and there are kind of some things forming in terms of the MCP protocol agent to agent protocols, everybody kind of understanding, okay, these LLMs are really good at receiving, processing and spitting out structured data. So now we have function and tool calling which didn't exist like two years ago in the way it does now.
So to answer the question of where everything is going is quite difficult.
[00:28:47] Speaker A: No, I think I'm aware. But you are certainly of all the guests I've had on the podcast, by far the most knowledgeable on AI.
Because my follow up question, and maybe this gives a little context in what I'm asking is like how difficult it is, how difficult is it to navigate where you're going to have to focus on?
[00:29:12] Speaker B: It's difficult because there are so many opportunities in the automatic context of the existing software and the friction points that people experience when using WordPress and when using WooCommerce that can be addressed through a lot of different implementations of AI. So you could essentially build solutions for so many surfaces of this automatic ecosystem in the dot com experience, in the WooCommerce experience, in writing content, in maintaining and managing and updating a website, in actually developing a WordPress website. So it's almost like where do we focus our energy and resources to be most impactful for the users of automatic products specifically? And that just ends up being a fun question to build and experiment and see where the users want us to go and see where the focuses of the business end up pointing and create products like the site builder and create the products that we're going to be releasing very soon and see how as we have more and more coverage of the ecosystem, how these products all work together and form this beautiful end to end system that's unified to that user and addresses all of their needs. So it's fun to try and keep up. It's difficult, especially in automatic where it's not just one single product, it's a lot of products that work together. And that same thinking applies to WordPress. It's one CMS, but it's a CMS that does anything because of the themes and the plugins and the blocks that come with it. So I'll leave it at that. And the content.
[00:30:51] Speaker A: Yes, I think it's. This is a, this could be very neatly summarized into. It's as much as living in the now can be forced through software.
[00:31:05] Speaker B: Yes. And it's also as much as living in the now can be forced through the lens of these multi billion dollar companies and massive user bases and need to solve user problems in a way where it's not super disruptive, but it's something that lays the foundation to pivot and evolve into this AI future of what the web will become. And that goes for Automattic in our products, but that also goes for WordPress and the project itself. Like how do you Keep relevant the WordPress project and the need for human publishing of content and presentation and digital expression and ensure that's protected and available and accessible into this. Whatever the future of the web becomes through all of the automatic positioning and the WordPress positioning.
[00:31:56] Speaker A: I agree it's almost an like three dimensional chess what you, what you're playing.
[00:32:05] Speaker B: And you know, I would probably say 99 dimensional chess.
[00:32:09] Speaker A: My point is that you're only seeing the two dimensional thing because, because that's where you can focus on, but you're aware there is multiple layers of where you need to also be looking and taken into account. It's. I, I can appreciate the complexity of what you're building and it is with that sort of enthusiasm that I saw the announcement yesterday on WordPress's 22nd birthday. Was that, was that picked particularly?
[00:32:39] Speaker B: No, it wasn't. It wasn't really planned. It kind of shaped up over the past several weeks. This has also been an AI team that I've been advocating for for a long time in terms of making sure the project at least has a surface for folks to share ideas and contribute and have the discussion about like where does this go into the future and how does this remain as impactful and beneficial over the next 22 years as it was over the past 22 years. So it really shaped up and it was announced on the birthday. Wasn't planned though.
[00:33:13] Speaker A: I kind of think it has to be, man, because if, if, if anything.
So there's, there's a whole bunch of teams, right? It's, it's the, the AI make team.
There's, there's a whole bunch of make teams that are important for the project.
Some of them have been disbanded, some of them are not getting the love they should or, or whatever.
And you can argue that there's a whole bunch of them are indeed pivotal.
But I think for me this is the, the one that is actually pivotal. So my favorite team is the WordPress core performance team, which is just part of core, but just to have a dedicated focus on performance for what is it, almost three years now, I think that's been a godsend. If you compare the stuff that's been added by that team and directed effort into every single release, making it more performant as we go from a sort of natural flow instead of just an afterthought which is I think the challenge for anything performance related to have that be part of your core thinking.
Every single thing I queried, am I querying it in the smartest way? Is my query somewhere crashed if I need to redo it within half a millisecond? Again, am I have to waste a lot of resource all of those types of thinking?
I love that that's part of, of core. Now what I love about the AI team is because, and I, I, like I said, I think it has to be kismet in, in, in one way that it's announced on a birthday because it means, for me it means this is what we recognize as something that's huge and not potentially pivotal, but it is pivotal because the moment we start introduc AI in a unified way and Felix's plugin AI service just scratches the surface because it's nothing more of providing a unified way of connecting your, whatever subscription you have, whether that's chatgpt or anthropic or whatever.
But that's the middleware part.
And on top of that we're going to have a concerted focus on trying to figure out everyone else's thoughts about AI and collectively pull them in. And I think for me this is a huge signal towards the, the health of this open source project. And I, I don't say project, I mean specifically the health of this open source project because I don't think there's many teams that have the potential as, as AI does. Because if you think about it, and I'm ranting here but, well, I'm agreeing.
[00:36:01] Speaker B: With everything that you're saying.
[00:36:02] Speaker A: Good.
Every single other team, literally every single other team that's part of the make.WordPress.org tree, plus all the sub teams that are part of it, every single one of them benefits from having a unified thought and approach of how AI is to thrive and work and not work and connect to the whole of WordPress. So that is the code base, that is the plugins, that is the theme, but it's every single API that's available to us. It's every single piece of content. When you're inside your WordPress site, like think about AI powered search inside your WordPress as well as on your WordPress site, all these types of, you know, huge potentials that you're introducing in this way.
I think, I don't think. Well, I wonder how many people see the magnitude of this team, this team's potential.
[00:37:06] Speaker B: Well, that was the main reason why I was pushing so hard for an AI team and this, this is something like I mentioned before, it didn't just happen in the past day. It's been something that, that has been planned and pushed for for a very long time and came to a place where it was time to actually form it and release it it right now. But I completely agree in terms of the actual surface of this team and what this team can do, but also what the team can facilitate in terms of the project and the thinking around, how do we develop in a more effective way, how do we handle these old tickets and bugs and communication.
[00:37:46] Speaker A: Thank you. I was going in there.
[00:37:48] Speaker B: Yeah. So it's not just to do with the actual core software, it's to do with the core project itself. And obviously there is the core bend of how do we make sure WordPress being so.
Being the scale that it is and containing the information and human expression that it does, how do we make sure that this is available and accessible in a smart way in an AI era, in the core sense. But also how do we make sure that, like you mentioned, the AI search and some of these other things, we ensure that core and maybe surrounding feature plugins and projects give developers the building blocks that they need to go and build WordPress into something that is incredibly impressive in this AI era. So it's a huge combination of so many different thoughts and ideas and concepts that we, as this founding team have and bring to the table, but also the founding team itself provides the surface for this community to ensure that WordPress is not just relevant, but incredibly successful and thriving into the future.
So the future is what I'm very focused on. How does WordPress remain something that it has been for the past 22 years and again, like, continue to grow and thrive into that future?
[00:39:04] Speaker A: Yeah, you mentioned something that was one of the first things I thought about, which is weird because I didn't mention in my monologue there, but the connection of WordPress track, GitHub in terms of all the listed issues and bugs and whatnot, which now is triaged by a human for the most part, and given the thousands, tens of thousands, I don't know how many tickets there are, it's close to impossible to solve everything.
But if you integrate all of that with AI in a, you know, in a fashion where it, where it starts doing its own reasoning and combining certain tickets into one and duplicating or. Look, this is an easy fix. There's no reason not to do this.
Or here's the solution. This ticket has been waiting for, but it got buried. And, you know, all of these Types of enrichments of the I'm reporting a bug. Let's hope it gets merged as a solution anytime soon. And there's tickets that are 12, 13, 14 years old, if not even older, but the potential just on that side of things so the quality of your software becomes just tremendously better combined with what that says in terms of health of the actual solution as it is. API enrichments easier triage for new stuff coming in like the the amount of stuff that can be built into the flow of just making WordPress as a project better. But specifically touching Core here, I don't think, I don't think we can fathom what this looks like in two years, let's call it that.
[00:41:03] Speaker B: I think there's absurd opportunity there and it's going to be very fun to experiment with the community and, and whatever ends up happening with all of that work. Imagine a world when or if there are no bug tickets or there are no maintenance things and the incredibly talented people that are occupied by that work now can think about the important things that WordPress does need as a core product.
So AI obviously has its implications in terms of making sure the core product is great, making sure that there are building blocks and feature plugins and APIs that support AI, but also making sure that the project itself is in an incredible spot to have the contributors contribute to what actually moves the the needle in terms of the success of the project as well, because it's not all AI. And I I actually this is something good to to mention. I've seen a lot of people thinking like oh great, an AI team, like they're just going to go and add ChatGPT everywhere in WordPress. That is not the approach. It's how do we make sure WordPress as a CMS remains effective. And honestly, that's probably not really many AI features in core. I don't expect there to be many, if any. It's more of like how can we allow people to build AI features and maintain kind of the way plugins have always worked on WordPress? You want a functionality, you find a plugin, you install the plugin or you build something yourself while also kind of coming around the back end and saying here's how the project can be impacted and improved and made to be more efficient. So that's the focus. We'll see where it goes. Because it's still early days and the point of teams is to get the community input and make sure we're doing something that is most impactful.
And those are going to be fun discussions over the next couple months for sure.
[00:42:56] Speaker A: Yeah, yeah. I mean, I don't think there's a word camp to be held where this is not going to get a major focus in terms of this. These are the conversations we're going to have, I would imagine. By the way, the plugin that I mentioned previously, AI Services from Felix Arns, I would imagine that becoming part of Core as a minimum.
[00:43:21] Speaker B: I think that's interesting. I think that's an interesting topic to unpack because we've never put these third party APIs and dependencies into the core project other than OEmbed.
So this actually pulls at some threads that will definitely be discussed over the next couple of months of what role do canonical plugins and feature plugins play in WordPress? What role do packages that enable developers play in WordPress? How can we make the core project more nimble?
[00:43:52] Speaker A: That's a, that's actually a discussion. That's. That introduces a whole extra layer that we've seen discussed over the last period since September, let's call it. That is the, the, the.
How much should we start thinking of WordPress in terms of like the Linux distribution model where you set up your Linux installation, you essentially get to determine which repo you're going to be using and a repo will have a, of components and that components you will either have or not have part of your WordPress site.
I think AI and certainly feature plugins as a whole.
I think it makes a lot of sense to start thinking into terms of a model where you can make those feature plugins more integrated into Core if you're not choosing certain feature plugins to become part of Core. For me personally, I think AI Services makes a lot of sense to have it in Core for the simple reason that it's going to. There's. There's so much that's going to be hooked into WordPress on all these different layers with regards to AI. It makes so much sense to unify that in one way in my opinion.
[00:45:08] Speaker B: I think what's also lovely is that Felix is the, the co lead of this group and we've actually put our heads together on these topics for, for years at this point back, back when we were in wpai, we were speaking together about what the plugins that we, the stuff we were building and the plugins that he was building could result in. So it's really fun to have Felix and Pascal and Jeff as these founding members of the team because every single one of them brings these projects and these thoughts to the table in terms of AI.
And what will be interesting is, I think regardless of what happens, all of this work will get closer to Core and will be explored through these canonical plugins in the same way that Performance did with the Performance Lab plugin and allow us to really like, see what works, what doesn't work, what does the community and the project actually need and want and ideally be like incredibly adaptive to WordPress as a whole and hopefully infuse quicker, newer, different thinking into the project, which is necessary in my opinion in this AI era where things are changing very rapidly. So it'll be fun to see how things shake out over the next couple months.
[00:46:21] Speaker A: Yeah, I mean me saying that I think AI services should be in Core is not like, let's merge it for, you know, six, nine. I think it's more of a. It makes a lot of sense for me to think in that direction because of the importance it has as a connecting agent.
I mean, like I said for in my head that this makes sense.
There's other solutions that I currently see in WordPress where I go like really then compared to those solutions and then what this allows us to do on a project scale specifically.
Yeah, like I said that this, this makes sense in. In my head, but I get that this is.
Adding anything structural to Core is always a very highly debated project because it's never as straightforward as oh, but this is still keeping us lean and mean. It's something we need. Sure, but have you thought about. And then, you know, all the people who definitely don't want to have another dependency inside of Core will go like, you know, we don't want it.
[00:47:29] Speaker B: Something that I am immediately excited about is just the fact that the AI services org gets exposed to more people because it's been out for a long time and a lot of people kind of know it it but not enough.
[00:47:40] Speaker A: I give it all the attention I can give it in my newsletter because I think it's just, it's awesome and.
[00:47:47] Speaker B: It'S really well built and it's built in terms of the eventual. Which has happened now, the eventual formation of this surface that can itself surface the AI services plugin and some of the MCP work that Pascal has done and some of the feature API work that we've done at Automatic can kind of just present this to a bigger collection of eyeballs.
[00:48:09] Speaker A: And what I like about all the experiments that you've done, that Pascal has done, they're shared on the respective GitHub profiles.
I think not too long ago you shared a whole bunch of libraries that on the automatics, GitHub account, I think.
[00:48:25] Speaker B: Yes. So the feature API and an MCP adapter for this feature API.
[00:48:31] Speaker A: Yeah. So those are cool tools.
I mean regardless of whether they're, they're shared and to play with, but they're fully open source as it's supposed to be, meaning everybody can learn from where you left off or vice versa.
And it, it almost for me feels like there's a little bit of an excitement getting back to understanding what the next phase of WordPress is going to be like.
One of, one of the things we talked about right before we hit record is something along the lines of you're half my age.
We're about, we're keeping it in mysteries.
But most certainly what I can say is that you have the age of my kids and the, your age bracket, if I'm just, you know, categorizing it as one, is a very different difficult category to get excited for the project.
I think there's, there's a, there's a, there's a whole bunch of reasons.
Christian Taylor asked me this question at Press Conf as well and I think it has a lot to do with how media is consumed for the very large portion of your age bracket and that's entirely video based and closed walls and all that. So open source and, and thinking along the lines of freedom is not necessarily on the forefront.
But what I'm going with this is that if you look at AI, roughly everybody young is excited about what AI does for them. They may not always get the whole potential there, but in terms of potential, what AI good and concerted focus for AI for youth is, I think it's a good entry into WordPress because it is powering the vast majority of the Internet. What is it, what are we going with now? 43%, whatever.
[00:50:41] Speaker B: I think we're at 43 right now.
[00:50:45] Speaker A: I think that is.
And I'd love to hear your thoughts on that specifically. But from my perspective, that is a huge opportunity that AI brings for the next generation.
[00:51:01] Speaker B: Yeah, well, I think there are so many ways to take the thought of infusing youth and excitement into WordPress again because you have the developer side and the ecosystem side of things where my generation of developers looks at WordPress and goes like, oh, I know WordPress. It's that like old CMS. And then you go like, well, I built a very large startup within WordPress and I got acquired and by the way, it powers half the Internet and they go like what? I never even knew that.
So you kind of have that side of Things where we need to figure out how we entice those bright people and the bright minds that go to the faang companies and go to different types of startups to this industry to make sure it thrives into the future. We also want to attract, and I don't think this is an age thing, but we want to attract the people who are entering into software because of AI who previously were.
There was never going to be the possibility that they enter in. And now AI makes it accessible for them to build and kind of showcase the fact that WordPress is this full stack way to build not just a website and not just a blog, but like a true application where everything's bundled into one thing. You have your auth, you have your capabilities, your users, etc and get people excited in that way. And then you also want to get people reinvigorated about the fact that personal ownership of data and presence on the Internet is very important, important. And it's probably going to become a lot more important as AI eats software and as AI gets infused into Instagram. And like, this is actually something I haven't really brought up or even thought about much, but there was an experiment in Facebook where they were building an Instagram, but it was all AI. And I don't know if it was added to Instagram or if it's just an experiment, but it's like the experiences on these closed platforms will become more and more led by AI. The X algorithm is using the underlying GROK AI stuff from what I hear. And all of that goes to show that there's not much control over these platforms and they're lovely places to be, but you still want the ability to control your presence and the information that you present on the web. And this comes in the form of websites right now, but it's going to come in the form of a repository of information about a user or more immersive experiences as AI allows people to be more expressive and creative. And WordPress needs to capitalize on that because if we don't, there's no other platform like what we have. No, there's no other project like what we have. And it is very clear that we could go down the path of this absurdly centralized web where it's not very fun and I don't like that. So I think that there are multiple aspects that we can pull on in terms of AI infusing excitement, bringing people in, in terms of developing, creating those who are developing, coming to the project because of what it has to offer, and also refocusing the Thought of this expression and ownership of data back on WordPress and away from closed gardens. And the way to start is to just start talking about it and do it in an official, formal way, which goes back to that AI tool team and what we can do there.
[00:54:25] Speaker A: I like that you said the, the category of people that otherwise wouldn't have found software.
The, the. We. We also know them as Vibe coders, I guess.
And I see a lot of people being kind of negative about Vibe coders. Like, you know, it's not secure, it doesn't scale and all that. Sure, there's a lot of validity in that, but I think what we're doing, kind of missing if we're approaching it from that direction is what you just said is that there's a whole group of people who are now being introduced to coding and whether that is on WordPress, with WordPress or even not, that previously, would have never even dared to look at code, play with code and find their way onto the Internet and thus becoming, call it a different version of a developer.
But most assuredly, those who are going to stick by it are going to find themselves becoming better and better as they learn, because the first thing they built was buggy as hell. What is the next step? Figure out how do I ask the AI for the bug checking and how do I learn from that? Okay, so now it does that. Let me remind myself for the next time I ask AI to build it, let me ask it to do it in this structured way. It just showed me while debugging it, like there's a whole, whole group of developers who will actually turn into a developer role just because they're approaching it with diligence, but from an entirely different route than just learning, literally starting with basic.
[00:55:55] Speaker B: And I've seen that in my industry of startups where previous business founders have pivoted to be a technical founder and create these initial MVPs and get them to a point where it becomes clear, obviously you need developers, but they're at the point where it's like, I've proved my idea like this works and now I can really scale it and really build it into something. And that goes outside of startups as well in terms of building websites right now, but also allowing people who want to be expressive but couldn't be without AI to paint on the web. Like, I think AI is a beautiful way to paint on the web and I would love to see the products that Automattic creates and the solutions that this AI team for.org create to open up that thing that doesn't really exist right now. Like, you can go on lovable, you can go on v0 and create interfaces and software, and that's awesome. But I want to get to a point where you can do that, but you can also create just whatever you're thinking in your brain and say, this is me online.
This is my business. This is my blog about what I actually like. And maybe a blog isn't just text anymore. It's maybe like.
I think a good example here is. I think it's James Kemp. James Kemp, a Woo product guy, is creating, like, these immersive web GL experiences where you can walk through galleries of the products that he's selling through WooCommerce. And that's how, like you would you say this to yourself three years ago, and you're like, there's no way. There's no way.
[00:57:28] Speaker A: Time.
[00:57:28] Speaker B: Yeah. And it's like this. This solution just comes out of creative expression. And I want our products at automatic and I want WordPress itself to be the place to go and do that. So, yeah, I think it's an exciting future if you allow it to be. And you need to strive towards the future that you want, because AI is coming and there's a lot of disruption that comes with AI and uncertainty, like we talked about. Like, even us building and the labs creating these models, models we're following, where they bring us.
So you want the voice and you want WordPress the project to be a good driving light in whatever happens with AI.
[00:58:12] Speaker A: 100% agree. And I think that's a lovely note to end on.
I was going to throw in there, let's paint the web. But you essentially even expanded on that more beautifully. Thank you so much, James. It's been quite insightful and lovely to have you on the podcast.
[00:58:26] Speaker B: Thank you very much, too. I always enjoy chatting with you and I hope to do it again.
[00:58:32] Speaker A: We shall.
Will you be at wordcamp Europe?
[00:58:35] Speaker B: I will be at wordcamp Europe and I am excited to see you and anybody else who's listening to this. If it comes out before.
About AI and WordPress.
[00:58:43] Speaker A: Yeah, probably not. I'll have to talk to Roman to make that happen over the weekend or something.
[00:58:48] Speaker B: Well, we'll also be at WordCamp US and WordCamp Asia and WordCamp EU. You're going to be everywhere and again and again and again. So if anybody sees me there, feel free to say hello.
[00:58:58] Speaker A: Cool. Thanks so much, man.
