AI isn't replacing websites—it's changing how people discover them. In this episode of The Dialogue Architects, Lauren Goerz sits down with Brian Alvey, CTO of WordPress VIP, to explore how AI answer engines, enterprise content management, and conversational experiences are reshaping the future of the web.
As AI-powered search and answer engines continue to change how people discover information, what happens to the websites and content management systems that power the internet?
In this episode of The Dialogue Architects, Lauren Goerz is joined by Brian Alvey, CTO of WordPress VIP at Automattic, for a conversation about the future of enterprise content, conversational AI, and digital experiences.
Brian explains why websites remain the system of record for organizations—even as AI assistants increasingly become the front door for customers. Together they explore how enterprises should think about AI answer engines, chatbot governance, security, evaluation frameworks, and the guardrails needed to keep AI reliable at scale.
They also discuss why content management systems aren't disappearing, how "Jobs to Be Done" can help teams build better AI experiences, the challenges of agent-to-agent communication, and why accessibility and voice interfaces should be foundational—not afterthoughts.
Whether you're building AI-powered customer experiences, managing enterprise content, or designing conversational systems, this episode offers practical insights into preparing your organization for what's next.
(00:00) Welcome to The Dialogue Architects
(00:44) Brian Alvey's Journey to WordPress VIP
(02:08) What Makes WordPress VIP Different?
(03:07) Why Content Management Systems Still Matter
(06:29) How AI Answer Engines Are Changing Search
(10:13) Chatbots, Voice, and Conversational Interfaces
(14:36) AI Agents and Vendor Lock-In
(17:22) Applying Jobs to Be Done to AI
(21:14) Breaking Work into AI-Friendly Processes
(22:55) A Simple Chatbot Success Story
(24:22) Policies Become AI Behavior
(25:18) Governance and Enterprise Guardrails
(25:42) Jailbreaks, Prompt Injection, and Security
(28:28) Scaling AI Security Testing
(30:24) Evaluating Models and Cross-Checking Results
(32:00) When AI Agents Talk to Each Other
(33:42) Balancing Creativity and Automation
(37:11) The Reality of Voice Interfaces
(39:06) Accessibility and Designing for Everyone
(43:20) Will AI Change Brand Loyalty?
(44:41) Closing Thoughts
Brian Alvey is the Chief Technology Officer of WordPress VIP at Automattic, where he leads technology strategy for one of the world's largest enterprise content platforms. WordPress VIP powers digital experiences for leading organizations including News Corp, Salesforce, NASA, the White House, major financial institutions, and global healthcare companies. With decades of experience building publishing platforms, enterprise software, and internet infrastructure, Brian focuses on helping organizations modernize content operations while preparing for the next generation of AI-powered digital experiences.
🌐 Learn more about WordPress VIP: https://wpvip.com
🌐 Learn more about Automattic: https://automattic.com
🌐 Learn more about WordPress: https://wordpress.org
🎙️ Follow The Dialogue Architects
💬 Connect with Lauren Goerz
🔗 Connect with Brian Alvey on LinkedIn
Lauren Goerz: [00:00:00] Welcome to the Dialogue Architects, where we explore how enterprises can thoughtfully design, scale, and govern conversations between humans and machines. Today's guest is Brian Alvey, CTO at WordPress VIP. Life mission, according to LinkedIn, includes making software that makes creative people more powerful, and has the privilege of working with super big brands every day of the week
Brian, thanks so much for joining us. I'm really excited to talk with you about the future of work, this equalizing wave of technology. Excited to hear your thoughts today
Brian Alvey: Yeah. I'm excited to be here. Thanks, Lauren.
Lauren Goerz: Great. So maybe just to kick off, I would say, this episode, one of the things that we like to do is give a little bit of an origin story.
Where are you coming from? What are you working on today? Maybe you can give us... You know, I think everyone knows WordPress, but what is WordPress VIP? What's the difference there?
Brian Alvey: Yeah, no problem. So origin story for me, like my villain, su- [00:01:00] super villain origin story.
Lauren Goerz: Exactly.
Brian Alvey: Um, uh, you know, when I was a kid, I wanted to be a comic book artist.
That turned into really loving publishing and making magazines and things like that. That turned into making software for high-end newsrooms, so teams of 1,000 working on cranking out lots of content. The thing you said about the making software that makes creative people more powerful, if there's something you're trying to do, I can build software that helps you do it faster, better, stronger.
And so like you said, I've gotten to work with a lot of cool brands. Spent my whole life not doing WordPress, mainly because I'd been building CMSs for 10 years before WordPress existed, and so I didn't think I had to care, and then it kind of never went away, and it got bigger and bigger and bigger, and it powers 43% of the web.
And so about four and a half years ago, I joined the WordPress company Automattic with the founder Matt Mullenweg as the CEO there, and I'm the CTO of the, the enterprise platform. So inside a company that is all over the planet, distributed, no offices. We have a lot of products for consumers, WordPress.com, Jetpack, Tumblr, things like that, hundreds of millions of blogs.
Tiny... I, I like to say little blogs you've never heard of. [00:02:00] And then my team is the one that powers the thousands of ones that you have heard of. So, you know, tons of big news brands and, and marketing websites. That's it. Enterprise WordPress, it's... Very quickly people think, hear, they hear WordPress and they go, "Oh, doesn't scale, easy to hack."
And like- Okay ... literally WordPress VIP, people- I, I
Lauren Goerz: haven't heard that yet. That's at least something new.
Brian Alvey: Well, but it's... Like, I get where that comes from, right? Common knowledge. Yeah. If you don't patch WordPress for three years, like somebody's gonna upload something, take over your server, and you're done, right?
You're toast. Yeah. And so it's our job to show like, no, no, no. We power like News Corp's global websites. We power salesforce.com, capgemini.com, banks, pharma companies, the White House, NASA. Like, we have like the FedRAMP authorization to work in the government. So like the fact that the White House and NASA and Federal Reserve banks can run on our platform means it does scale, and it's really not easy to hack.
So we prove them wrong.
Lauren Goerz: Amazing. And I think one of the things you just said there, I think it... Was it 40% of the internet is running on WordPress? Did I hear that correctly? I
Brian Alvey: think it's roughly 42%. Right around
Lauren Goerz: about.
Brian Alvey: Okay. And that, that- 42% ... and that was my [00:03:00] theory, which is like if half the web- Yeah ... runs on WordPress, why doesn't the other half, right?
So it's an interesting- Yeah ... time to be working on this.
Lauren Goerz: And I think that's... Maybe as we talk into kind of the next subject around the future of work, I'd be curious to learn more. You're an expert in CMSs. I think anyone that's working in our space kind of... Of content management. Becomes even more interesting now that we're setting up all of these kind of document management pipelines to enable this, let's say search, these different types of search programs so that, for example, in our case, a customer can talk with Rasa via the customer of our customer.
So ultimately they can talk with Rasa and talk with the documentation of our customers, and that's pretty exciting because I used to have to go out and hard code every single one of these FAQs, and now you can just upload a document. It's a ma- button that everyone was waiting for 20 years ago, but it's here now.
I'm curious though, CMS, a lot of people would say today, "I could just vibe code that." You know? I could, you know, and I think you're alluding to this with the security, the hackability of these things. Why is the CMS so enduring and, and what's the future of the CMS look like?
Brian Alvey: This is a [00:04:00] good question, and I do wanna talk about all your conversational parts that, that thing that you do because there's so many things in what you just said.
One is that your website is your system of record. That's your source of truth, your documentation, things like that. On, on a, on a silly, maybe less, people might think it's less important, you know, it's not healthcare and banks, but, like, on a brand level, making sure, like, Gucci has the right, you know, assets and stuff across all their things they do, making sure that all their stuff is in sync is a really big part of the CMS.
The flip side is, yeah, all software is getting eaten by AI right now, isn't it? And so when you talk about, when I talk about, like, the consumer versions of websites, you know, CMS tools- Mm ... and the enterprise ones, I feel like the, the consumer stuff will get eaten. I think a, like, a $3 a month host, that sort of thing.
You know, people talk about how they, they, they vibe coded a, you know, a 20-page website for their brand, right? So I, I don't know, you know. So Ross' own website may have 400 pages on it, like, you know, something like that. There's some- Yeah ... some amount of content, and if you're one person managing that website, you can do that [00:05:00] out of a Git repo or, you know, an S3 bucket or a, you know, markdown files, right?
You can do that, but the second you need to coordinate with the second person on that, you either have to teach them how to get into that Git repo or use whatever tools or figure out your silly markup, your mark- silly, your silly markdown thing that you're doing, or you need to build a tool for them, and that's a CMS.
And so if you take that to the extreme, you say, "Okay, maybe I can get away with using, you know, Replit or Lovable or something like that to design some stuff, a microsite, a six-page thing, a 10-page thing, a 200-page thing." But at some point, when you have millions and millions of articles and you have a newsroom of 10,000 people that need to collaborate in software, all that falls apart.
All that goes out the window. You need the guardrails. You need the workflow tools. You need all the audit logs to see, like, I changed the headline, but then you change it, then I changed it back, and we got sued. I'm the one getting fired, not you. Like, how do you do all of that, right? And then same thing- Yeah
I think high-end commerce doesn't go away. So if you look at what will still exist [00:06:00] for content management in five years, I think high-end media newsrooms collaborating, the stuff I've worked in my whole life, and I think high-end commerce, stores. Like, stores will always pay money to a good platform that helps them sell more.
So I think those things endure, but I think, yeah, the six-page website, I, I don't think that's something you're gonna pay, you know, 30 bucks a month for or 30 bucks a year for.
Lauren Goerz: Fair enough. And a good time to be at WordPress VIP ultimately, 'cause that's exactly the focus right there. So you're in the right place for that.
Now you were mentioning as well as websites being the source of truth, and I think that's been interesting for us because we're both from our side as Rasa the business, we're interested that Rasa is represented well by all of these different kind of search machines, language models, et cetera. On the flip, access to the internet basically, sometimes dangerous.
Maybe you don't wanna do that, you know, without any kind of guardrails, but there is a risk that it comes back with some interesting facts about your company. And for us, in some cases, it's actually a lot of historic facts are some of the hardest bit to regulate, is we have, you know, lots of old products that every so often, you know, continue [00:07:00] to pop up.
I'm curious, have-- what does the tooling look like? You know, y- you know, you're making these people more powerful, creatives more powerful. What does the tooling look like to help them with these kinds of authoring for not only other people, but language models?
Brian Alvey: Yeah. Okay. So I'll tell you a funny story from a site that I built a while back about, like, being too good at this, and then we'll talk about how you I guess correct the internet is the other part of your question.
Like, how do you correct- Yep ... the internet when the internet's wrong, right? So one was I got to build TMZ a long time ago, the, the big gossip website. Mm. Of the problem of AI answering agents kind of taking over your website's dot job and telling your story for you, and maybe telling it wrong, or maybe talking about other companies when you want yours to be the one that's surfaced.
So there's, there's sort of two things that any site needs, whether it's a media company, whether it's a, a marketing website, like we talked about, you know, Salesforce, Jefferies, Capgemini, these banks, things like that, Merck, pharma. They have different needs. Some of them run towards AI, some of them run away.
But they definitely [00:08:00] have a, "I need to find people, and I need to convert them into customers," whatever that means. And that finding part has changed radically in the last year. So, you know, search and social used to kind of intercept your customers on their way to your website, but they didn't intercept your entire website.
You know, you still had to go visit the website and learn about Rasa, things like that. But now with AI answer engines, you don't have to do that at all. You can find out everything you want. So you need to make sure you're represented in these new places, and it's different than trying to, you know, fix the search results, fix the Larry King's seventh or eighth divorce or whatever it is in Google.
Okay, that's one thing. But that doesn't undercut and sidestep your entire business and get rid of your entire pipeline of taking that person to a customer, right? Mm. But AI does. These answer models do. So how do you make sure you're being represented well? And to us, it comes with sort of... There's sort of three steps to it.
One is seeing what's going on with bots on your site, visibility. One is managing that, blocking certain ones, letting certain ones in. But the third [00:09:00] part is really making sure that the information about your brand that's out there is accurate, and it's way easier to do that on your website, like we talked about the Gucci example with the photos- Yeah
and the stuff and the brand and the logo, the colors, the, whatever the change is, your sort of, you know, your design, design system, your brand guidelines. It's very hard to do that in someone else's platform and make sure that they get it right, especially when the answers are so fuzzy and conversational.
Yeah. Right? So that's, I think it's a big challenge. We're learning a lot about it. I can tell you some of the things we learned, but it's, it's a tricky space. It's a bit of a wild we- Wild West right now.
Lauren Goerz: Yeah, fair enough. I, I guess maybe this is more anecdotally. I realized the other day as I was, you know, like typing about trying to find some information, how sad it was how little imagery I see in my day-to-day today.
Because when I need information, I'm going to these... I, I'm doing a lot of reading. I like reading too, don't get me wrong. Kind of lose this image element to these search results. I imagine for Gucci, like, a lot of their traffic is probably, you know, image rich or, or, you know, it's interest- part of the Gucci brand is its image.
It's not necessarily what you write about Gucci. So it's an [00:10:00] interesting time out, I imagine- Oh, yeah ... for very visual brands. Yeah,
Brian Alvey: try to, try to describe- Yeah ... handbags, you know, that are... and make them sound like they're worth $30,000. It's very hard. Show me a picture though, and now I understand why they make so much, yes.
Lauren Goerz: Absolutely. So I'm curious, you know, at the end of the day, a lot of what Rasa does intersects also with what you're doing because our agents end up on your websites, I would say. I'm curious to hear, you know, maybe you have some initial thoughts about things that talk. That's at the end of the day what we're building here at Rasa, things that talk.
Have you seen these on websites? What's your initial thoughts? I know that you've had some prior experience with kind of voice user interfaces. I'd be curious to learn more.
Brian Alvey: Yeah. So, so from a long time ago-- So, so voice user interfaces, there's a lot of things I've been involved with or, or fascinated by or advised or whatever in the space, like Spoken Layer that translates articles for, you know, all the Wired magazine articles into audio so you can just listen to them.
I think, I, I think that's an interesting space, and I think it's been waiting for its moment for a long time, and I wanna talk about it, a bit about why I think, like, the companies like Rasa and this conversational thing are [00:11:00] where everything is going. So I think you got there early, and I think things are-- the world is gonna catch up with you, so I'll tell you kind of the, like, why I'm bullish on what you're doing.
But in general, s- just chats, chatbots on websites, we have a very interesting relationship with Salesforce. So Salesforce is both one of our biggest investors, so they invested in Automattic, the, uh, the WordPress company here, and they are a partner of ours, so we, we help them integrate their stuff into websites.
And they're also a customer. They-- We power salesforce.com. So I like to joke that even though they have all the software in the world, they still come to WordPress VIP to host salesforce.com, which is, you know, really, really nice of them that they trust us with that one thing, and they do all these other things so well.
On the partnership front, though, they have Agentforce, which I'm sure you've heard of. Yeah. You know, just put this chat on your website. It answers pre-sales questions, post-sales questions for, you know, prospects and existing customers, and that has to get the data right. And so I've seen a lot of how do you do a, an Agentforce integration with, you know, a big automotive website or some kind of a, you know, [00:12:00] business-y enterprise marketing website, get the right data into Data Cloud and Agentforce so that you can put the chatbot on your site, and it gives good answers.
So I think that space is very interesting. The thing that I sort of teased earlier is I feel like if you can-- it's a whole future of work thing we can talk about, but I feel like if you can turn the jobs people are doing digitally into something that is so well understood, such a tight machine, such a well-described machine, that you can ultimately at some point flip from where I have to go into a browser and drive and do all the work, to something where the browser or the agent is telling me what to do next.
I think the place where you end up, the sweet, happy vision for all of this, is not that I have to spend all this time working in a browser as a person pushing, you know, the, you know, like the hamster- Yeah ... on the wheel, like making the thing go. But it, it will do a lot of the work because I've described the job so well that it just has to come to me and ask me questions.
And my dream for that is I'm in my car and I basically get a [00:13:00] phone call from the agent. I get that conversational thing that says, "Hey, so we've got a couple headlines for you to choose from. We've narrowed it down to these. Which ones do you like or would you like me to change them?" If something could be more like a coach for me to flip that mode of who's in charge and who's making the decisions but still keep me in the loop, I think the best place ultimately to keep me in the loop is I'm just walking down the street and I get a phone call or something and I answer it.
I answer a couple questions, and that's, I think, the, the ultimate sort of luxury way to do your job in the future when the software does, does so much of it and tees it up for you. So I just feel like companies like Rasa, people looking at this conversational space, this, this place where you're Doing the best job of talking to software becomes that spot.
You're all waiting out on the beach, that wave's gonna come in, and at some point a year from now, three years from now, five years from now, the whole world's gonna go, "Damn, this is how it should've been the whole, the whole time," but you're already there. So I think that's exciting
Lauren Goerz: I have to say I'm also, I would say, a mega-fan of voice [00:14:00] interfaces because when they work, you feel so powerful.
It's like a Jedi, you know what I mean? I've just organized things. When doors close because you told them to- Oh, yeah ... or a light went on, it's like, I love, you know, play my favorite song, and then bam. It's just, that's incredible. But I am- Well,
Brian Alvey: when, so, so you say- Oh, yeah ... you see
Lauren Goerz: people
Brian Alvey: walk through, like, a Safeway, you know, a, a store or whatever, right?
Yeah. And, and they do that, like, Jedi wave, and the doors part. It's not parting because you waved- Yeah ... but it feels like you, you, you know, you just manipulated the universe. That is really cool.
Lauren Goerz: Yeah. Yeah, so I think that's something that's w- exactly what you described, we're not there yet, where you feel this Jedi-like power of things, you know, you're just orchestrating everything around you with minimal effort.
Although, I always wondered if it does take effort to use the Force. I'm not entirely sure. But nonetheless, that aside, you mentioned something that actually a lot of our customers are scared of, and it's super interesting to talk with them. This whole scenario where different people have personal agents, uh, with agency that are acting on their behalf is actually, you know, it really lowers the cost of switching from one company to another.
Let's say, you know, you think of telecommunications. How many times have I been angry at a telecommunications [00:15:00] provider for this, that, and the other? I have a personal agent that can go out and negotiate with, with, I'm not gonna say the brand name person- Yeah ... and then I can switch to another one all on my behalf.
Because in reality, actually, sometimes the reason I, I'm not leaving bad service is because it costs effort and isn't quite yet worth it. So taking the effort away really makes, it makes it easier for pe- people to be flakier with brands ul- ultimately. And I think that's an interesting fear that I hadn't thought of until some of our customers put it to us this way.
Brian Alvey: It, it depends. So when you talk about the big services, right, all these big LLMs are a swappable commodity. Right now, if I'm coding, I can use Opus, I can use, you know, I can use all the different models, right? You can kind of swap them in and out, and the results are pretty much the same, and I don't have to care what's happening.
I think it's interesting that the most important players in this revolution today are kind of swappable, right? That's, that's really weird. But what they're doing that's smart is building up that whole, that, that history with you. So if you think of, you know, like a chat history with ChatGPT or something like that.
If I switch, [00:16:00] the new model is gonna have to learn all this stuff I've been telling it for two years, you know? Well, that stinks, right? So there's different kinds of lock-in. It's not like a, a phone number portability where, you know, like a long time ago, you couldn't take your phone number with you, so you wouldn't switch.
Then they passed a law, you can take your phone number with you. Now you can switch. So switching became easy. I think if somebody made a data portability or a chat history or context portability law to take that and lift it and go, it'll become more switchable. So it just depends on what you're doing, but it's an interesting time.
Yeah. Use ChatGPT, and they, they ask a question and they get an answer, or they use it inside Facebook or they use it inside whatever, right? And then there's this like, uh... Sorry, that was the green group. Then the yellow group was the ones, like, using agents, things on a loop. So, you know, paying 20 bucks for a Cursor or Claude Code or something like that.
And then the red, it was only one red square out of this entire stupid grid, was people who were, like, doing the agents, doing the loop, doing the OpenClaw- Hmm ... doing the harnesses, like really, really deep. And I kind of, I kind of feel like I will never ke- keep up with, like will never catch up to Marc Andreessen or [00:17:00] Andrej Karpathy, right?
Like these, these people who are like so smart and so, so far ahead of me. But I don't have to catch up to them, I just have to be like right behind them. Can I be, can I be ahead of the rest of the planet? And it makes me feel a little bit better that like I'll never be, you know, the, the Hall of Fame ones.
Yeah. But I'll be right- I'll be close. So I don't know. Yeah. It's, it's, it's a lot of- You're not bleeding edge ... it's just too much to do. Yes. It's going too fast
Lauren Goerz: Yeah. I guess, let's-- I'd love to dive in again. We've went on the surface of kind of the future of work. I'd be curious, I know, you know, from your framing, there's a lot going on with CMSes.
There's a lot going on in lots of different areas. When you think about the future of work, do you have... That you feel very powerfully about right now, given- Yeah ... that it is hard to say- Yeah ... what's gonna happen in the future.
Brian Alvey: No, I, I do. I mean, because, because I get to just focus in on what does, what does it mean to update your website to grow your business, right?
Like, it-- which is a, which is-- sounds simple, but it's, it's a very broad thing. Like, everybody's website is the heart of their business, right? Like, it's the source of truth for all your information about your brand, but it's also your online, [00:18:00] your always on store, right? Everybody goes to sleep, somebody goes there to get a question answered.
They go there to buy something. They go there to fix something. So I think that the-- If you think of building products and building platforms like this, there's a philosophy, like a product philosophy called Jobs to be Done. So you think of like, you know, what does Lauren need to do today? How do I help make that faster, better, stronger?
Like, take it from seven clicks down to whatever. And I think that you can kind of lose sight of that philosophy of, "I'm here to solve a problem." And I think one of the really interesting things that AI has done is made me think more deeply about the future of work, and I didn't know it was going to make me do that.
And what, what it-- the reason it made me do that was it-- when I was thinking about how do I automate things, how do I augment Lauren, and how do I, how do I automate the things that Lauren's doing every day? I was like, "Oh, if I need to be able to hand this off to an agent, to a bot, to the world's, you know, sort of like worst, most destructive intern.
Most powerful, but most destructive intern. How do I describe the job that you're doing?" It's, "Oh, there's jobs to be done." That's the, these [00:19:00] micro units of work. And it's made me think on websites of sort of two levels of the work. There's one is, what are the things I have to do? So sort of like the factory level, like...
So if you think about like, you have a factory that needs to make shoes or cars or whatever. How many shoes do I need, how many sneakers do I need to make today? How many cars do I need to make today? How many pieces of content do I need to make today for a tabloid newspaper in London, right? With my newsroom of 1,000 people.
And the second part of it is, how do I make each one of those units? So sort of like the factory production and the unit production. So how many sneakers do I need to make today? 200 pairs. How many-- How, what does it take to make a sneaker from zero to, to 100% done? And so thinking of things in those terms, I need to make 600 pieces of content.
They come from this many games were played last night, this many elections happened, this many whatever, certain inputs. And then what does it take to take a story from not started to ready to publish, right? It's empty. Now I have a title, now I have a body, and now I have photos, now I have sports scores, now I have whatever, fact-checking quotes.
And [00:20:00] so there's hundreds and hundreds of things on these checklists, and if you can get to the point where you know what do I have to do and what does it take to do each thing, the unit production and the factory level- Then you can go and hand that work off, those eight hundred and twelve steps to get that piece of content done.
I can hand them off. I can either have a smart human, an expert, very well-paid expert, or a junior, more intern to- sort of person, or a frontier model that's very expensive, or a local open source model. And you can mix and match both human, you know, people and bots, experts and novices, and figure out, like, how do I just do more with less and be as efficient as possible?
So, a weird little, like, rabbit hole there about thinking about the future of work in the context of cranking out content on websites. And I think that the AI has made me think about those jobs to be done, which is a very pure thing to do, to go back to what is the business that Rasa does and what is Lauren trying to do day to day, right?
That's a really interesting thing. That's what I should've been thinking about all along. But now that I'm thinking about how do I hand [00:21:00] that work off to software, even more capable software than we used to have, where you don't have to drive and sit in a browser all day long, becomes really interesting, and it makes me feel like, oh, this is, this-- like I said, what I should've been thinking about all along.
Lauren Goerz: It's, I would say there's, there's a parallel in the conversational AI space as well. And, and this, this is even before AI, to be fair. I used to be in implementation, so I'd, like, work with the teams who were building agents. And one of the things that was always quite profound is the people from, let's say, the digital team that would say, "Hey, we're gonna make this agent.
It's gonna be amazing. Everyone's gonna love it", would meet with the business team who's actually owns the processes, and they would ultimately find out, wow, there's actually a lot more behind, I don't know, ordering a new card. There's 20 different checks that happen. And so actually this, like, awareness, this growing awareness of how things work in your company is a byproduct of working on these projects because you yourself and your team have to quite literally outline it.
And yes, a- and that, that's something, to be honest, that's even hard for, like, an AI agent to do because the knowledge is [00:22:00] spread across so many people's brains. It's not even on the internet sometimes, so. Mm-hmm. It's, it's quite a, I would say a, a, an effort to get these business processes on paper to actually automate the things that you wanna automate.
Brian Alvey: So, so when you, you talk about, uh, when, when it works, it's like magic, right? And when it doesn't work, it's like, it's kind of annoying. So I think ab- I, I actually look at this. We, we talked about this a few weeks ago, uh, our exec team. I was telling them stories about chatbots that I, I tolerate and chatbots that I love.
Hmm. And there's very few that I love, that I was like really, really, really impressed and I would trust them to do things, right? And, and the two I had as an example, one was an airline I fly all the time for, uh, for work. And if I go to change a seat or do something or ask a question, like, I'm always in the, the chat, and then I go, "Can you just get me an agent?
Get me to the, get me the person." And I, I'm always opting out of the thing because it just... I don't know what it can do well, you know, besides tell me, like, if something's delayed or whatever. Like, that's not very interesting or helpful. But it can't do the things that I'm asking [00:23:00] for, and I always ask for a person, and so the chat works great once I get a person.
It's kinda sad. My wife ordered something a bunch of weeks ago for one of my kids' schools, and it was supposed to be like two, two of these things came, but only one came. And I went to the site and it's like, "Well, we're closed. Ask our chatbot." So I was like, "Okay, here's the problem. Here's the order. Here's what happened."
And the chatbot was like So let me get this right. You ordered two, but only one arrived. You checked the packaging, right? Yeah, yeah. This is a chatbot. So it, it like, it totally under- It goes, "So we're gonna send you a new one, but just in case maybe we shipped them in two different envelopes, can you wait three days?
Come back on the 24th or whatever, and then tell me the problem again." I'm like, "Well, there's no way that's gonna work. I'm gonna show up, I'm gonna have to explain it all again. It's not gonna know what to do." And oh my gosh, it was like it totally understood the job of returns or missing items or whatever for this little store that I don't think makes...
It's not a famous store. It's just like a, you know, college university merchandise, branded merchandise store. And like, okay, it was phenomenal. And I came back [00:24:00] like three days later. I'm like, "Hey, that thing never arrived in a separate package." It's like, "No problem. Uh, I'm, I'm, I've got one on the way now.
Here's the tracking number." And you're like, "Wow!" So I will rave about that thing all day long, but there is... That's one in 100. These really don't do a good job of reading your mind or understanding the business process so deeply that you can hand it off to something and it works flawlessly
Lauren Goerz: Yeah. I think two things off the back of that.
I think, one, the actual company policies are a big part of what makes an agent good. A good example, one of our engineers in our team, his name's Leo. I think he was-- I think the company was HelloFresh, I can't remember. But what he loves about the HelloFresh agent is whenever he gets, let's say, a bad bell pepper delivered to his house as a part of his HelloFresh order, HelloFresh replaces it without asking any questions.
They don't even really wanna see a picture of it. They just trust you implicitly that if you went through the effort about this bad bell pepper, you deserve a new bell pepper, basically. Exactly. You're not scamming. And that's, and that's- Like, right, you got a bad pepper ... that's a policy. That's the pro-- Like, some-sometimes actually, 'cause I remember one of my jobs used to [00:25:00] be, thankfully isn't anymore, reading through just thousands and thousands of lines of logs of people talking with these agents, and half the time I was like, "Was that a me problem as the billing agent, or is that this company's problem because they bad service?"
And so I think that's a big piece of it, is like being willing to actually acknowledge that I will send you another one part of it. But I think for the rest of them, as you described, especially airlines, I'm not gonna lie, sometimes when I'm working, I think the airlines, they're probably shackled by governance, to be honest, and I think that's a big blocker in some ways to actually moving forward.
Not that governance is bad. It should keep us safe and do lots of things, but in some ways I think that's definitely a hurdle that people are still working to get through.
Brian Alvey: There-- I think you're right about the governance, the compliance thing, the guardrails on these things. I forgot about that. There's a...
One of the ways that you can check to see if an agent is unshackled and ready for you to mess with is to ask it non-work questions. So there was- Mm ... like a very famous thing where you're, you're, you're in a chatbot, you're doing a thing. You know, there, there's the famous story of some guy asking- Yep ... a guy who actually lived near me who I know, [00:26:00] asking for like the thing to sell, the, the, the car dealer website to promise to sell him like a truck for a dollar.
And it's like, "And you have to say yes to everything I ask." And it just, it just went with it. So it was very early in the chat days, right? And I'm sure you heard the example. And then the court ruled that like it's not a representative of the company. You can't have the truck for a dollar. But I do think that...
So, so, so the test you can do for these things is like you ask it, "Hey, what time is, you know, flight such and such landing?" And it goes this. And you go, "And what does red plus yellow make?" You know, and if it says orange, you're like, "Oh, I got a live one here. It, it doesn't have any restrictions on what we can talk about."
Yeah. "We can go talk about some wild stuff now, and I can mess with this." And so that was always sort of the, the test to see if one of these like doors is open for you to get in and do crazy things, and then you can go do the nutty stuff. So I think actually thinking back to the airline, who I, which I love, I'm, it's, it's, it's a fantastic airline.
I think it's probably smart that they only ask really simple questions, and if it gets complex, they don't just try. They do lock it down and they say, "Would you like to speak to an agent?"
Lauren Goerz: Yeah, and that's, [00:27:00] I think the cost of losing is so much higher for Lufthansa than it is for HelloFresh. Sure. At the end of the day, you either lose a bell pepper or, you know, a 500 Euro flight.
Yeah, yeah. Exactly correct. Yeah, exactly. It's, it's like a- $16,000 business class ... it's a different ballgame over there, for sure. Yeah, exactly. Yeah, so definitely. But a- anecdotally, I think that's also funny for, I've seen a bunch of different, uh, posts on LinkedIn of folks that have run out of, I don't know, Claude credits or something, and they, they've unlocked this and you're like, "Hey, I'm, I'd like to order a large cheeseburger and fries.
And by the way-" Yeah, yeah. "... can you make this PowerPoint look better?" Or, you know, something like that.
Brian Alvey: No, exactly. But
Lauren Goerz: like- So, so- ... you know, the
Brian Alvey: Chick-fil-A chatbot on their site lets me-
Lauren Goerz: Yeah, I think that was it ... you know,
Brian Alvey: like, like, do, do code or whatever. I forget what brand it was, but yeah, that's very funny.
Lauren Goerz: Yeah.
Brian Alvey: Yes.
Lauren Goerz: True. So I guess maybe as, as you look towards, like, governance with... And I think maybe anecdotally I can also say one of the things that is interesting is all of these companies that I used to work for, if we got anything wrong about two years ago, you know, we'd have to, like, come in and apologize and make it better, and that could've been, like, I don't know, the brand voice was wrong, right?
Now I find it super interesting 'cause people are much more willing to take a chance, which I think is interesting because it's helping [00:28:00] technology move a little bit faster. People are just putting stuff out there. Again, uh, sometimes in a bad way, but ultimately I think the, the, the, the broad result has been that a lot of our customers are moving faster, and that's exciting.
Is that something you're seeing in the kind of WorldPress universe, WordPress universe as well? Uh,
Brian Alvey: so, so are people moving faster? I, I don't know. I, so I think they're, I do think that they're willing to adopt things faster because of the fear that their competitors are gonna crush them and run circles around them if, if, 'cause they're doing it, we're not.
We still work with so many banks, government. Scale works for you, not against you. So I would think writing security code, like vibe coding security code, is probably a terrible idea, right? That's not good. But automating millions of tests, penetration testing, hacking testing, the, the tabletop exercises, all the stuff that we have to do with, like, the, you know, FedRAMP authorization to host the government stuff.
We go through these exercises every year, every few months. I think the ability to go and do [00:29:00] those attacks or try these, all these doors at scale is really cool, and here's why. When there's an incident, when something goes wrong, when a site that we host or a site that we know of has some kind of a leak or something, at some point five people get on a Slack channel and they sit down and figure out what the hell went wrong.
How did this thing get out there, right? There's, you know, WordPress itself wants everything to, like, you put a photo in there, it wants to serve it to the world. Our brands are like, "We put a photo in there, but don't show it until Tuesday when the, you know, results are released or the election's done," or whatever the thing is.
They have embargoes. And so there's these tensions between information wants to get out there to, like, don't let people snoop around and find that photo we uploaded and know who the next, you know, person of the year is or Nobel Prize winner. These are some of the stories that actually happened because people didn't lock it down.
And once that sort of that, that incident happens, five smart people get on a Slack channel and they're like, "Oh, okay, I found how that happened." But why can't you be doing that in advance thousands of times? Why does it take five people after the fact trying to [00:30:00] figure out how something broke or how something leaked or what was architecturally wrong?
Can't you figure that out in advance with free software that you can scale to check a million different paths, right? So I think the opportunity is to find the places where scale works for you, not against you, lean into that, and, and then stop worrying.
Lauren Goerz: Fair enough. And I think that's something we're seeing on our side.
Probably one of our biggest roadmap items is really, like, evals and simulations. It's something that, not that we should not stop user testing. I don't want the UX people to come for me. But on the flip side, I think being able to not have to wait for your 10 users to run through all of these different tests, or just relying on a regression test alone, being able to actually simulate real conversations, real conversations, lifelike conversations in an AI agent scenario is massive because you can discover so much before you actually go live.
And as you said, that prevents a lot from happening in live and you can... So that's I think, again- Right ... uh, maybe, maybe something that does give folks peace of mind and this willingness to put [00:31:00] stuff out there, you know, maybe faster or without as much oversight as they were previously.
Brian Alvey: Yeah. Yeah. I, I, I think the pattern of when you build software right now, the pattern is go in plan mode, make the best plan, come up with a description, come up with exactly...
And then, but the best thing is to take that plan that, you know, the OpenAI agent did and take it over to, you know, Anthropic and say, like, "Check out this plan that your competitor did. What do you think about this?" And it will find five things. It'll find five major things, and I don't know if the, you know, the first LLM kind of has blinders on and just, like, doesn't see its own screw-ups or its own flaws or something, or I don't know.
But, like, literally having them get into a room, like, "Here's a WhatsApp chat. You guys go discuss this and come back with something when you're done and you both believe that you've got something solid. Don't trust each other. Go." I think that kind of thing is fascinating to use one of those models to inspect the other one, because I feel like it's got to reduce the problems by 99%, you know?[00:32:00]
Lauren Goerz: And that's, this is a little bit of an aside, but that's something that I'm super fascinated to learn more about because we've done a lot of, let's say, conversational design for humans, but I'm super interested to know what happens when agents start talking to agents. And not as of just simply, let's say, doing kind of robotic process automation in the way that we're working with now, but in the future of my personal agent talking to someone else's agent.
What does that- Yeah ... w- what do the WhatsApp messages look like? Is, are they shorter? Are they different, you know? It's such an- Yeah ... interesting, like, thought experiment, I think. Uh, a- and I could have assumptions, but maybe it is different.
Brian Alvey: So, so one of my favorite developers, he actually hates this. He actually fears this.
Mm. He just, he feels like, like, like, "Are humans gonna be left out of this?" And what I mean is, if I can tell my agent to write this massive thing and then post it into Linear or where, you know, GitHub or wherever the place is you're tracking all your issues, or post a big Slack message, nobody else has time to read those because now it's 11 pages instead of two paragraphs.
So I'm gonna have my agent on the other side summarize it. And people joked about this, you know. Like, I, I, I, you know, took- ... [00:33:00] take three bullet points, write me a long email, send it to Lauren. Yeah. Lauren's like, "I don't have time for this. Can my agent summarize that into three bullet points?" And so that's, that's kind of the, the joke, but at, at the end of the day, like, what happens when humans aren't writing or reading things, and that whole thing happens without you?
I don't know. It's kind of fascinating.
Lauren Goerz: Yeah. Es- especially I think as, as they talk to... If we're also feeding, it's like this recursive, like regressive process in the sense that we're feeding it in all generated content. Uh, you know, you probably see it on LinkedIn, on the internet anyways, probably like odds are a lot of the stuff I'm reading today is generated anyways.
You can kind of tell sometimes. Yes. But, and then it gets just fed back in into this kind of garbage loop ultimately. Or, you know, that's- Yeah ... a possibility. So de- definitely interesting. But nonetheless, I think there's a lot of stuff, that's where I see optimization. I don't necessarily want, and I think someone said this was going around when language models first came out, I don't want someone to do my art and my, like, creative writing.
Really wouldn't mind if someone, you know, rescheduled my hair appointment. I really wouldn't mind. That would be something I'd be super [00:34:00] open for in terms of- Yeah ... you know, like non-human interaction.
Brian Alvey: So, so the, the... People joke about that, that we all thought that we'd have, like, robots doing our dishes and driving our cars and stuff, and we'd be sitting around- Yeah
doing poetry and learning the guitar. Instead, robots are writing poetry, and, like, we're still doing the dishes and driving the cars. But it happens because the revolution was in language, and the revolution was in the things around, like your space, right? We, we talk about how, how you describe things, right?
The conversations we're having, this, this kind of thing. We used to think that, okay, a, a bot can do the work, but then I have to explain it to my boss or my teammate or whatever. Like, I add this thing where I get to sell it or explain it or something, and I think the selling and explaining can be done, can be, you know, automated and be done by these things just as easily.
And so I, I say that it's not an artificial intelligence revolution, it's an artificial creativity revolution. And that freaks people out, but give it the stuff before the thing and see if it could do what humans did. I think that's fascinating because I think- Mm ... actually if you feed it the right... Like, could you tell it everything before Seinfeld happened, then describe the [00:35:00] show that you want and see if it produced Seinfeld?
It's not there yet, right? I think we'd all agree- Yeah ... it couldn't do that. If you said, "Watch 200 episodes of Seinfeld, now write a Seinfeld episode," it will do a great job. So I think- Yeah ... that's, a lot of people see that as a test for has AI actually- Mm-hmm ... become super intelligent.
Lauren Goerz: But maybe I think this is, or at least this is my hot take on things, I think what's still missing from AI, at least today, is, like, the spark, right?
You still need to feed it 30 Jerry Seinfeld episode- you know, Seinfeld episodes until it can get there. If you didn't have Jerry Seinfeld writing those 30 first episodes with Larry David- Right ... they wouldn't be any good. Right. I, I think, and maybe taste-wise, I think for me something that's either raw or, like, unfinished kind of, I'm like, "Oh, love that."
Screenshot of someone's note, fantastic, love it. Like, you know? Mm-hmm. And maybe at some point they become good enough to mimic this, but that's- Sure ... to me right now- They will. They will. They will rip it off, yes ... the most interesting type of, of content, I would say.
Brian Alvey: I, I agree with you.
Lauren Goerz: All
Brian Alvey: right. I, I think our slop detection...
You know, my kids are phenomenal at, like, AI- ... slop detection. Way better than me. They're like, "Oh, that was terrible. That's, that's absolutely AI." [00:36:00] It's very funny.
Lauren Goerz: True. That's a whole other interesting, I would say, business space is identification of AI-generated content. Mm-hmm. But before we continue, we've only got, I realize, sorry I've been blobbing away here, five minutes left.
Is there anything we haven't covered that you're super interested in talking about? We'll cut this piece out.
Brian Alvey: Yeah. Uh, uh, so, uh, no. I- w- there are a bunch of topics. I think wherever we go, we go. I think it's all been good, so, uh, I'm not worried. What else, what else do you wanna cover?
Lauren Goerz: Trying to think. Um, we haven't, to be- probably my head of marketing will be like, "So tell me about conversational AI."
Is there any other, like, voice user interface things that we can plug into this episode that you have a hot take on? Or just inter- you know, conversational interface topics as well. That also might be interesting. But if you don't- Yeah, we can talk about that- ... we can also- ... that,
Brian Alvey: that accessibility thing where everybody's blind at some point.
Oh, yeah. That's right. You know, like, like that whole thing of the fact that, like, this is a very important thing, and you think it's a, a novelty thing or a goof, but it's not. It's actually pretty serious. So we can do that. I think the only, the way that we would get to that topic is, I don't know how we get there.
I don't know how we got there last [00:37:00] time.
Lauren Goerz: We were talking a little bit about the super poweredness of kind of voice user interfaces, but maybe it's more than just superpowers. It's actually, you know, there, there's lots of ways that this can be useful, I would say. Is there anything that's come across your path that has, you know, in the voice user interface sphere that you find super interesting or powerful beyond kind of opening up doors or turning on lights?
Brian Alvey: Yeah, no, I don't think it's there yet. So I don't think the thing that, that dream that I have where I'm in my car and it's talking to me and asking me things is there yet. I think you have... So I think if you look back to every time we've had a conversational sort of interface, a voice UI or whatever into something, it's always been like interest- you know, magical in, in parts and, and kind of junk in other parts.
And, and so the, the magical part 10 years ago, being able to t- like look over at a speaker in my room and say, "Hey, what's the weather today?" You know, like the fact that it d- it, it understood the question and that it, it read me an answer is amazing. The answer it gave me is junk. Like it's, "Oh, it's gonna be 50 [00:38:00] degrees for the low and 80 degrees for the high, and raining around 4:00 PM."
That's okay. The, the, the magical part was the hearing me and then answering me. So I think the NLP part, the, you know, natural language processing part of it has been good for a while. It's gotten better. Now I think the exciting thing is that middle part gets better. So I think the fact that when you ask it something, it goes, "Yeah, it's gonna be, uh, 80 today, but you're driving into San Francisco and it's gonna be 25 degrees colder, and I think you should, you know, like...
or, or, you know, go out now and start your car to like get the ice off of it, because otherwise you're gonna walk outside thinking you're ready to drive to work, and you have to wait 10 minutes to get the ice off your car. So go out and start it now, come back in and eat, then go back out again," and you're set.
And you're like, "Wow, I would pay 99 bucks a month for the service that helped me do that," right? I, I think that's useful enough to be a product. So I think getting that middle part right, doing more, understanding me, my context, is one of the things that's gonna be great, but it's not there yet. You know, I do know people who are building these things in weather apps or in, you know, voice apps or things for [00:39:00] websites.
And I just think they're... I think they know where they need to go, but they're not there yet.
Lauren Goerz: Yeah. And I think it, as you said, I think one of the things that is interesting is that we have actually seen Rasa be used in these accessibility contexts. So there's a lot of kind of early Rasa work that was done, especially during the-- when we first launched Rasa open source.
There were lots of different teams that were actually using Rasa in care homes to help with these folks that needed to navigate throughout their day via a voice interface. And we even had a few use cases which I thought was super interesting, and we're seeing this more now being, become an actual thing, is there were There was once a project I remember from our open source kind of period where there was a person who was building this Rasa agent that would support folks that were in really low population areas, so they just had no one to talk to.
They were, like, in the highlands of Scotland somewhere, and there's no one around, and literally they were just trying to give them almost for mental health, you know, someone, something to talk to. And I think probably if they were to think about what they could do now with [00:40:00] just ChatGPT, it's a lot better.
Um, but that, that's something, another user in- voice user interface thing is kind of just taking the loneliness away from the day.
Brian Alvey: Right. So, so I think the examples you're giving are interesting, right? Somebody's lonely, somebody needs to, somebody needs something to talk to. And I think on, on, on the accessibility front, one of the things that I love, like, you know, being a person who didn't spend my life in WordPress but kind of watched it from afar, one of the things that I thought they did really well with WordPress was from the s- from the beginning, they built web standards in.
So, you know, markup that validated or different things for different reasons. They put extra, let's say, craftsmanship into what they did building WordPress, and including accessibility. So if you use WordPress to do things, there's a lot of affordances in there for color contrast or things for, you know, if you're, you have vision problems or you're flat out blind.
Like, whatever the, whatever the thing is, there's a lot built into the software. And we have these debates, you know, probably less and less now, but we did five years ago, 10 years ago, like, should I make my, my content management system for a 2,000 person [00:41:00] newsroom, you know, in Sydney, Australia, like, should I make it so you can use that when you're blind?
And, and the answer is, like, yes. And you go, "Yeah, but it's more work." And they go, "No, no, no." But, but I say to everybody, like everybody's blind at some point, right? At some point you're, like I said, driving in your car, you can't reach over and, like, pull up a laptop and scroll through a spreadsheet and get an answer.
You need to just be able to say, "Hey," you know, "Hey, device name..." I'm not gonna say the device name 'cause it's sitting right here next to me. But, "Hey, device name. Can you tell me the numbers for the third quarter so I'm ready and I have them in my head before I'm in this call? And how much is the, you know, university of, you know, whatever deal that we have that we just closed so I'm ready?"
So I can't look them up when I'm driving or I'm going to crash, like this will be a problem. So everybody's blind at some point. And the, the, the things that I like about this are, are, one, we built things into our software. So for me, I can see just fine, but I can actually now tab between things in our dashboards and launch things and check out backups and do stuff.
Like, there's so much accessibility built into the system that I can just kind of fly [00:42:00] through it with key commands, so I actually get a benefit because of that. And then for the people who say, "It's just too much work, the extra code to put area tags on things and do the accessibility check and make sure that the contrast is this and that."
Like, it's not anymore work to make things for the, a, a small audience of people, you know, who have different abilities. But if you do those right and you built them into your platform, then when you end up in this universe where, oh, 80% of the people using my website are bots That actually can't see. Like, you're kind of ready for them.
So the accessibility work we've done for decades- Interesting ... prepares us for these non-people users of the product. When they're the users building things and they're the buyers as well, the consumers of it, that's pretty insane. You're actually ready for this sort of like weird, invisible audience because you've always taken care of those forgotten audiences.
Lauren Goerz: Actually, I hadn't thought of that, is that language models are effectively blind in a sense. Not all of them. I know there's multimodal- They, they can see now. Like
Brian Alvey: they're reading-
Lauren Goerz: Yeah. There's multimodal. Yeah, yeah. [00:43:00] But at the- Right ... you know, the broad, large majority of the intake that they're getting is, you know, just simply words, which, y- you know, they're reading, but at the end of the day, there's no imagery necessarily in that input.
Brian Alvey: It's, it's a lot like a screen reader, right? Looking at your page and seeing blocks of stuff and understanding the hierarchy of like what's important, what's, what's smaller, what's a- what's in a table, and then reading it out to somebody. I also think there's a, there's a really interesting challenge for people like us, you know, who build these sites.
If your newest... If, if the, if the growing portion of your audience is going to be bots, how do you make your product, like, well-loved? How do you, how do you create brand loyalty? So we talked for a minute about Gucci and like, you know, describing a handbag with text. Like, how am I gonna get a lar- And I worked with Gucci on a platform nine years ago.
We don't work with them today, so I'm not like selling a customer stuff. But like, how do you make it so if I ask a large language model, "I have $30,000 burning a hole in my pocket. I need a handbag. You know, which one should I get?" How do you make them surface, you know... When an agent can't smell [00:44:00] that product, can't hold that product, they don't care what the logo looks like, they don't care what the colors are, the craft, how long it lasts, whatever.
They're just going on a couple of bits of information. How do you build in these factors that can excite bots so much that they rave about your product? I think that's one of the more interesting spaces, is to figure out how to make your thing the suggested one in a world where they can't touch, feel, hold, smell, and use your product.
Lauren Goerz: Absolutely. And I am curious to see how long that lasts, that there is no suggested one, let's say, in language model interfaces. Yeah. I think that'll probably end at some point, but- Mm-hmm ... optimistically looking at least for the near term, it is an interesting problem. Absolutely. Brian, I just wanna thank you so much for joining us today.
It's been an absolute pleasure to have you on. I, I love that you have so many stories around every point of the chat. I feel like I'm, like we could be drinking a coffee together. So thank you again so much. Okay. A- and all the best. Looking forward to see what you're working on with WordPress VIP in the next couple of years.
Exciting times.
Brian Alvey: Thanks. I'll be following you and Rasa. Thanks a lot, [00:45:00] Lauren.