Talking Drupal #540 - Acquia Source

John: This is talking Drupal, a weekly chat about web design development from a group of people with one thing in common. We love Drupal. This is episode five 40, aquia Source. On today's show, we're talking about Aquias, fully managed Drupal sas, Acquia source, what it can do, what you can do with it, and how it could change your organization With our guests, Matthew, we'll also cover AI single page importer as our module of.

Welcome to Talking Drupal. Our guest today is Matthew Grasmick. Matthew is Acquis VP of innovation, collaborating across teams to shape Acquis product vision for artificial intelligence. He's a longtime Aquarion since 2012 and Drupal list since 2009. Having LED products to mark to market, including Acquia Source, Acquia, CLI Cloud, IDE Code Studio migrate, accelerate lightning pipelines, node js, and cloud actions.

He's also led Acquia Drupal Acceleration team to contribute heavily to Drupal nine and 10 major releases, project browser, automa, automatic updates, layout builder media workflow, J-S-O-N-A-P-I and more. Matthew, it feels like you've done it all. Welcome back to the show and thanks for joining us.

Matthew: Yeah, sorry about how long that, that intro was.

I'm gonna trim that next time. Some of those are old projects, but it's, it's great to be back.

John: That's a Okay. I just wish I did a better job reading it.

Nic: I, I just wanna know, are, are you an aqua Aquarius or,

Matthew: Aqui is typically how I pronounce it. I don't, is

John: that I did say aquarion, huh?

Matthew: Yeah, but that's,

John: that's

Matthew: okay.

John: Aian. Yeah, we should, yeah. We'll just edit that in later. Or, or maybe not you.

Matthew: No, I like it.

John: And for those of you that probably do already know, 'cause you've listened to me bumble through the intro before, or maybe you're new to the show and this is your first time. I'm John Ozzi solutions architect at eam.

And today my co-hosts are Katherine Buki, backend developer, architect at Mindcraft. How's it going? Thumbs up. She gave me a thumbs up. Perfect. And Nick Laughlin, founder at Enlightened Development.

Nic: Happy to be here.

John: And now to talk about our module of the week. Let's turn it over to

Martins. What do you have this.

Martin: Thanks John. Have you ever wanted to use AI to help map various content items on an existing site to structured fields on a Drupal site as part of creating a node? There's a module for that. It's called AI single page Importer, and was created in January of 20 20 26, so less than a month ago by Mark Conroy, who listeners may know from his work on the local gov distribution and install profile.

It has a 1.0 0.0 alpha three version available where it's works with Drupal core 10 or 11. It is actively maintained and for documentation, it has a pretty extensive remi, which is also currently. Now it has no issues yet and is officially in use by two sites according to drip.org, which probably isn't surprising considering how new it's now.

With this module enabled, you'll have a new AI content import section at the top of the node creation form. In there, you can provide the URL of the existing page to use and then click import content with ai. That will trigger a process where OpenAI will ingest and analyze. The existing page will extract values to populate your node fields, and then you can review or change those values.

Before saving the node in the configuration, you can specify the AI model to use a maximum content length and HTP request timeout value, which content types should have the importer available, and then abuse by specifying block domains, a flood limit and a flood window. You'll also need to grant. Use the importer for any user that should have access.

In the configuration, you can specify the AI model to use a maximum content length and HDTP request timeout value, which content types should have the import available, and then also prevent abuse by specifying block domains, a flood limit and a flood window. You will also need to grant a new user permission to use the importer for any user rules that should have.

Access. Module also includes a number of safeguards. For example, it will only accept URLs using HTTP or HTTP S protocols. Private IP ranges are blocked and by default it will only allow five requests per user per hour will perform HTML purification for long's text fields and strip out tags for short text fields.

In addition, it removes dangerous attributes like on click or inline JavaScript and generates CK editor compatible output. It currently supports a list of field types that include text, long text with summary string, text, date, time, date range, and timestamps. And link fields. It also supports entity reference fields, but only for taxonomy terms.

Now, listeners may also be aware of the unstructured module, which does some similar things, but requires you to use an unstructured service or run a server using their software. So I would say that AI single page importer is perhaps a little more narrow in scope, but works with an open AI account instead of requiring the less commonly used unstructured.

So John, I believe you nominated this module. So, is this one that you've actually had a chance to use yourself? Ha

John: ha, I have not. I saw it and thought it was interesting and decided to drop it into, into the module of the week to, to learn more. So I'm glad, I'm glad the process works. Yeah, I mean, I just thought it was, I just thought it was an interesting use case for, for AI and importers and, and to be able to kind of take content from various places and drop it into your, into your Drupal site.

It could be something that maybe I use in the future. I'm not, you know, I, I do have feeds running on my site currently to pull in subsequently talking Drupal episodes. But yeah, maybe this is a, a replacement for that or a different approach to, to bringing that content into, into my website.

I dunno,

Nic: I, I'll say I, I personally have kind of mixed feelings about this kind of thing. 'cause on the one hand. It feels like it does get back to kind of the root of one of the things that makes Drupal great, right? Or made dral great in the seven days, which was the ease of like integrating with different systems and pulling stuff in, like feed tamper, all those things made it very easy for, to pull, pull stuff in.

Mm-hmm.

The I'm a little concerned on,

by the same token, I'm a little concerned about the ease that it makes it to scrape other people's sites because we already have, are having ma we already have massive problems with bots in general. AI bots are scraping the internet or making the world a worse place. The internet worse, and it's creating a huge burden on my clients to block them or make them act like, responsible citizens on the net and making it so just any person can turn their website into an AI scraper I think is problematic.

So I, I have mixed feelings because it's, it's like, I think it can be helpful, but it can also be something that can, can cause problems for other sites, other sites, whether they intend to or not. Unstructured is slightly different. I, I have a client that uses it fairly extensively. It's more, it's more about pulling content from particular documents.

Although I suppose you could kind of print a page to PDF and then scrape it. They say they use ai. I'm not convinced that they do. I think that they, it's one of those services that just puts AI on it because everybody puts AI on it. At its core, it's really an OCR system. It feels kind of crazy to me to really, truly use AI for that, maybe as like a second pass or something if they couldn't understand something.

But yeah, unstructured has been fairly, fairly rock solid for us, but they've been fairly happy with it.

John: So, I don't know, like, I feel like, I feel like the use case for this module, like again, I'm thinking of my own use case that I just, I just highlighted, right? Like, just 'cause this uses, you know, uses ai, I mean, it, it's using it to analyze the content and help you dynamically drop it into the right fields, I'm assuming.

Right? But one thing that, you know, our talking Drupal feed doesn't. Actually have in it, I don't believe is our cover art, right? So I'm able to pull all the relevant content into my website and, and a link out to the show on, on our, on, on the Talking Drupal website. But it doesn't, it doesn't get that cover art.

So like if I were able to point it at talking Drupal and just get the cover art and get all the other other stuff and, and maybe less of the other information I don't need, like, I, I feel like, I feel like that's a good, a good use case. I mean, I think to your point, Nick, like whether, whether this is using AI to do it, to make, to make the, the person who's using its life easier or whether somebody's just gonna go and try to scrape content, like that's gonna happen whether this module exists or not.

So, I mean, I dunno, I like the fact that the AI makes it a little bit easier to, to kind of like, map that content for me.

Martin: So I, I think I should maybe jump in with a couple of quick comments. So number one, this module currently does. Does not support bringing over images which is interesting. I, I feel like that is actually one of the things that I've seen unstructured demo to be able to do.

So that actually I think is a difference between them right now.

John: Interesting.

Martin: Yeah, in terms of use cases, I guess when I was looking at this module, the, the first use cases that came to mind for me were like doing kind of a small site migration. So if you had like a 10 or 20 page site that was currently on WordPress or something like that, this could sort of make that process probably significantly faster.

Or the other use case might be, let's say you were in the early stages of a Drupal site build and wanted to demo some of the content types that you had built. This could be an easy way to pull existing sites or existing content out of a site that's built on another platform into the, the content types that you've built as a way to sort of demo that to the customer while a different team was working on, let's say, you know, a full site migration in the background or something like that.

So I, I hadn't necessarily. Gone. AI bots are consuming the world. But I think there are probably valid use cases where this could be used in, you know, fairly you know, non nefarious ways. I suppose.

Catherine: I went where Nick was going, oh, I'm sorry, Matthew, were you trying to say something?

Matthew: Yeah, but I'll, I'll wait for you.

Catherine: Yeah, I mean, I went where Nick was going and I, I, I do. I do think it's an issue. I, I also understand what you guys are saying. There's, there's part of me that is a bit like, look, this is, we're going where we're going, right? Yeah. This, this train's not gonna stop. So I, I still think we need to have these conversations about, you know, I'm not particularly thrilled about it being easier for bots to scrape more content and for this kind of cyclical pattern of, of content sharing or content reuse to happen across multiple websites.

And that, that's, I think we need to have this conversation, the conversation that Nick has, has brought up. But I think we need to have it for reasons that are, are, I'm not really sure, end up anywhere. Right. I still think we need to have it, 'cause hopefully it ends up somewhere. But things are moving so quickly and things are changing so fast and the train is moving so quickly that I, I don't see, I don't see where it goes.

We still have to have it. We still have to talk about these things, but I don't know what the end result of it is.

Matthew: I agree. The cat's outta the bag on this. Like that's, you're not gonna stop bots, you're not gonna stop scraping. And this school I'd actually, I have a different comment here. So I use AI all the time. If I wanted to migrate something, if I wanted to import content, I would not use this tool, use this tool.

I, I, I think this is an old way of working, so I'm sorry to be critical, but the old way of working was building a module and doing things inside of Drupal. And this, I think even the idea of having agents in Drupal I question it. I, I think Drupal's really good at a lot of things and we should use it for those things.

And I think already other things are better at agent orchestration and agent management, and that's gonna continue. Triple's not gonna win in that race. And the. I think this is gonna influence and hopefully this is something that we'll see at DrupalCon Chicago and the future strategy for Drupal and the Drupal AI initiative, but the strategic approach of Drupal ai where we focus on building ways to create tools for agents to use to reach into Dral from the outside and skills to have the right context.

Because with the right context and the right tools, you know, you can use Codex or Cloud Code or Cowork, and you'd say, please just migrate this to dral. And it would know, okay, here are the endpoints that are available. Here's how I authenticate. There's no, you don't need to worry about does this module support images.

You know, you, you don't need to build a capability for it to choose, even with ai, the right field types that can be done and can. Inherently because you'll expose the content model to an agent that's more general purpose and will understand how to modify the content model if necessary. So, and, and actually, I mean, this will dovetail later if we talk about source or, or other things and the role of AI in it.

But this is a kind of recent paradigm shift that I think not everyone has maybe even heard of. Certainly not caught onto. The, the new emerging standards for tool use and skills use and the new tools that are available, like the, the experimental feature that was released last week with cloud code having teams or open AI frontier where you just, you can't keep up with that level of orchestration inside of A CMS.

You've gotta kind of seed that. The orchestration battle has been lost to the, the big AI players here. And the question is, how do you, how do you fit yourself into the ecosystem? How do you, how does Drupal fit in that ecosystem? And how, what is its value at that point? And there's still a lot of things that's really valuable.

Like it's, you know, the access controls great, the structured content is great. The built-in workflows are great, and at least I'll try to wrap up my comic here, but like the, the value I found recently in using Drupal. Source and other things is like, it is a nice, I know the edges of the sandbox that I can put the agent into and what it can break and what it can't break.

So I don't, I'm not vibe coding, access control. I'm not vibe coding roles and permissions and workloads and structure. I can trust that Drupal's done that. It owns it. I'm not gonna break it. I can trust that the agent's gonna play within the bounds of that well established system. And that's the value of Drupal in that world.

And the better it can play within the bounds of it, the more likely I am to use it. So the value of Drupal to me is directly correlated to how well exposed those inherent capabilities are. But for that reason, I would never consider using an AI importer module or really any of the AI modules inside of Drupal at this point.

Catherine: Are you playing or are you, are you talking a bit about like MCP servers? I'm thinking it the, I, I'm thinking about the last few months I've had to build out a few MCP servers exposing Drupal's. Content structure model to Yeah,

Matthew: absolutely.

Catherine: You're talking about that? Yeah. Yeah.

Matthew: The Drupal AI module, the Drupal AI initiative is focused on the AI module.

I think they need to shift focus on the, the m CCP module, which Mateo, I forget his last name. Seeing it's a small side project right now.

Catherine: Yeah.

Matthew: Because currently if Drupal wants to discover the content model or like certain metadata, certain configuration mm-hmm. What's the workflow? What are the transitions, what are the states to get at that it needs direct database access or DR.

Access or some kind of workaround, you know, can read config on the file system but it can't do it from across a network. Yeah. And that just inhibits it so much. Yeah. But it has is JSON API and that pretty powerful. You can go pretty far, but it, it forces you to do everything the wrong way if that's the only tool you have.

Catherine: It's been fine doing the MCP service stuff, to be honest, and I haven't done a ton of Anyway. You're rolling

Matthew: over with, with something. I dunno.

John: I, I, he is muted so we Oh, oh, okay. It, I think he wants us, I think he wants us to move on.

Nic: No, no, no. At the risk of complete derailing this and may, maybe this is for Catherine's talk in, in a couple of weeks, but hearing that just makes my skin crawl.

Like the, the

Matthew: Maybe it should too. Maybe it should.

Nic: Like, yeah. The, the, and there's two parts to it. One is from a security and a data sanitization side, like giving direct database access to these models. I don't trust these companies and they've shown that they're not trustworthy. So like exposing that data, I mean, you're just opening yourself, like you're opening yourself up for liability, right?

Like set aside the quality control, that kinda stuff. Like if you. Providing these types of changes in production, which you, you need to, if it's like a data import, right? Unless you're changing the model of Drupal and you're saying like, do it on a dev site and then just input the database. Yeah.

Matthew: I don't think you have to give them production access.

The way I've, I've set up AI workflows lately where it has sandbox access, it sets up its own idea, AWS infrastructure, right? Like I, I ha it creates a sandbox. I approve the sandbox, it runs in the sandbox, and it uses telegram to send me a message and I hit approve or deny. But when I hit, approve or deny, it doesn't go back to the ai, it goes to a separate deterministic system with a web hook that decides to copy things over.

So actually the agent's in a sandbox where it never has access to do anything to production even wanted, okay. And there's a lot of things you can do like that, where you create an air gap, but that's, it's, you want the agent to run free within your defined constraints.

Nic: Okay?

Matthew: If it's not very valuable.

Nic: That's much more, more reasonable. The other piece that, the other piece, I'm, I'm struggling to really put this to words, but it's something that's been formulated over the last few weeks.

Okay. So I, I've been struggling to formulate this May, maybe it's a blog post or something, but I'm seeing a ai, like, yeah, AI has a lot of hype behind it. It's solving problems, it's also creating significant problems, and there's this, something just seems wrong when the solution to the problems that AI is creating seems to always be, let's find a way to apply AI slightly differently.

Rather than having AI solve those problems inherently, I guess like one of the problems that GI GitHub right now is dealing with is, you know, a lot of open source projects are just being slammed under a tsunami of just vibe coded lop, right? What, whether you think AI can produce good results, the fact that it enables so much output that has no value is a problem.

And a lot of the solutions I'm seeing are just like, oh, we need to provide an AI tool to review the AI results and get rid of them. Rather, like, but it's, I, I don't see how that, how applying more of the thing that's causing the problem can ultimately solve problem.

Matthew: So, lemme, lemme use a metaphor here, or actually an analogy.

Imagine if you, instead of ai, you're talking about humans. Humans cause all the problems. And we keep trying to throw humans at the problems to solve the problems. We put access controls around them, processes around them, and humans keep making a bunch of slop. Like a lot of it is the same inherent problems, right?

You create something that acts like intelligence, whether it is intelligent or not, is a, is a different story. But I, I think that that's it. Like you have to kind of treat it that way. And if you talk about, like, look at a social network and you give people access to the internet and the kind of stuff they publish that's awful.

And you're like, God, let's let's disconnect them. Let's get humans off the system. Let's not network it together. We're just putting more trash on top of trash. I mean, it's the same paradigm, really. The difference is higher volume, more speed, you know, like we're just doing it at this accelerated rate because it's simulated, but it is still the best tool for the job.

And that you, that's the thing, you're not gonna get away from the fact that it's the best tool for the job and it's becoming better at it. Day. So no one's ever gonna choose another tool long term. Well,

Nic: okay. I, I would, before we get back to the actual show, I would just say I don't necessarily know I, yes, that it might help some things, but I think in a lot of cases I think one problem is that a lot of the people that we have on the show are experts in the field and do a lot of putting those guardrails.

I think in general for sure. I think for the, I think for the general populace, it's not that AI is the best tool for the job, for most things that people are using it for. I think that it's the easiest tool for the job.

John: Oh yeah, absolutely.

Nic: And I,

John: hundred

Nic: percent. And I think, I think that's the problem. I don't even think, I think most of the time for most things that people are using AI for, it's the worst tool for the job, but because it's the easiest people turn to it.

And so I think, and I think that's part of the problem, like the same, same thing for like. Yes, there is a company out there that's finding real actual security holes with ai, and they're vet, they're vetting them and their numbers are showing that 30% of only 30% of the reported ones by their tour are actually real.

But they have actual experts looking at them and verifying them before they post them. But. Does that mean that AI is the best tool? Like it's just identifying things. You can use static analysis once you start to identify those patterns. And

Catherine: so you want humans to be less lazy. That's what you're saying.

John: We,

Nic: we, we can get back to the show. Now.

John: If you replace, if you replace AI to Matthew's point, if you replace AI with any, any other tool or term or thing, like, I think you're probably, you're probably gonna come to the same sort of, same sort of realization that the problem here isn't really necessarily the, the tool or the ai, the problem is the people using the tool or the ai.

Not to say that as, as people that are building these tools and these ais, right? Like we should, we should be you know, we should be trying to, trying to think of the, this stuff and, and trying to, trying to safeguard folks from, from these, these, you know, the dangerous use cases. But with that, I'd like to just say, Martin, look what you've done.

That's

Nic: what you wanted to say.

John: And I'd also like to thank you for sparking this insightful conversation with your module of the week. If folks wanted to suggest a module that could send us on another tangent or just chat with you about, about Drup stuff, how could they do that?

Martin: We are always happy to talk about other projects that could be future discussion grenades for other episodes in the Talking Drupal channel of Drupal Slack.

Or folks can reach out to me directly as Ben Clue on all of the Drupal and social platforms.

John: Awesome. Thanks Martin.

Martin: Thanks all.

Nic: Thanks.

John: All right, so, we are talking about Aqueous source. I'm sure we're gonna, we're gonna veer over into that AI topic again at some point. But for the time being, matthew, can you kind of give us a little insight as to the, the inspiration behind Acquia source?

Matthew: Yeah. It's a lot to say. I, I think it, the most fun and interesting way to start is to talk about something that started maybe 15 or more years ago at the very beginning of Acquia actually, which is a product called Drupal Gardens.

Mm. If you remember that.

Nic: Yep.

Matthew: It was actually really successful. It hundreds of thousands of sites grew very rapidly on it, and eventually Acquia basically replaced it with something called Site Factory.

Catherine: Mm-hmm.

Matthew: And the reasons for that were at the time. Everybody wanted deeper customization. They didn't like the, the guardrails of sa, they said, I, I've got millions of dollars to spend.

I'm gonna hire professional services. My, my web app is gonna be my identity as a company. It's gonna be our flagship of innovation.

Catherine: Yeah.

Matthew: We can't have guardrails or, or blockers on the features that we're allowed to have. And so, we created Site Factory to say, all right, this is scalable. And customizable.

I think the world has changed. Technology has changed. Websites, web apps in particular, software more generally, it's commoditized. People don't wanna pay as much for it. I think that the people are beginning to value ease and low low effort and low cost over feature richness. They're willing to make trade offs.

And you know what, what we heard from customers was yes, we do wanna be able to have an option to be super flexible and customizable, but also we have a lot of sites that don't need that, where we just want the cost lower. We, we don't wanna maintain it, we don't wanna update it. We don't have the Drupal experts available, or if we do, we don't want them to spend their time on that.

There's more useful things they could spend their time on.

Catherine: Mm-hmm.

Matthew: And it wasn't a very good solution for them. So we kind of came full circle and was like, well, we need something like Drupal Gardens, but how can we do it a bit differently so we don't box people in in the same way that Drupal Gardens did, it was very simplistic.

So that inspiration, right? This is a customer problem and pain and, and history. That's the inspiration. It's also what inspired a lot of the architectural directions of Drupal Canvas. Because mm-hmm. When we started imagining Aquia sources, when we're like, well, we need a totally different page building solution 'cause.

Drupal doesn't have good front end and backend separation. Put aside the headless thing for a second. Like you gotta, the front end developers have to go into the PHP and they gotta write pre-process functions. And the config is in the ui, but it's also deployed and you gotta put it in Git. So it's really hard for SaaS because everything is mixed together, together.

And so Drupal Canvas did a couple things. It it has a modern experience that's just much nicer for, for authors. Like that was a ne necessary thing. It uses modern languages, react mainly, but it's, it creates a separation of backend and front end while giving maximum flexibility on the front end. So you can write completely custom react code and you don't need access to the backend.

And I, I'm not sure you're familiar, you're with it, but actually if you wanna pull in entities or something, some kind of dynamic thing in Drupal Canvas, it actually uses JSON API from React. To do it. So good thing we invested in that API first thing. So that's what you know, brought about source.

We built it. Another thing we said though was we, we don't, we don't ever want to limit what's great about dral. We need to always embrace it. We can't we stop people. We don't wanna stop people from using the power and flexibility of it. So we put an export button on it and we said, you know, if you want, you can use source to get started fast.

Pilot something, build something small. It's not intended to be custom to, to the nth degree, but if you change your mind, hit export, and then it's a custom Drupal app. You can host it with us, you can host it somewhere else. You can take full control and you can treat it like an accelerator and that way you still own your code and, and data and that sort of thing.

And yeah, it's gone pretty well, although I would say it's still nascent. It's still new. You know, we have some customers that are very successful on it, but it is, we want it to be more feature rich. And the thing holding it back is the same thing that holds Drupal CMS or Drupal Canvas back. It's like we're unlocking those gates.

We're unlocking multilingual support for Drupal Canvas, for instance, is a big one second that's in Drupal Canvas. It's an aquia source, but it's kind of advancing at the pace of canvas right now. And, and most of Canvas is developed by Aquia.

Nic: So I, I guess the big question I always have with AA products then is how, how much does it cost?

Because you said that it brings the cost way down. Almost every service for Aqua requires a pretty in-depth sales process. Is Aquia source a commodity can, is it just like, hey, it's 50 bucks a month and you get two sites, or what is the cost?

Matthew: It's a lot less expensive like a aquia doesn't like publish costs that you buy off the shelf really anymore.

Like you'd have to talk to a salesperson. I don't think that's always necessarily gonna be true. So I can't give you an exact price because the price kind of depends. But what I can definitely say is it's much less expensive intentionally. And I think that we will be able, because of the, the structure of SaaS to offer it like, you know, direct kind of B B2C direct to an individual who can buy it for a flat rate off of the website.

It kind of lends itself to that much more than our, our products have in the past.

Nic: Okay, so, so you can't gimme a range. It's not like, is that hundreds of dollars a month? Is it tens of dollars a month?

Matthew: If it's, so if you do it per site per month, it's not in the thousands. It's, you know, definitely in the hundreds.

And I think it's even like the low-ish hundreds. Okay. But it, it really depends. There's economies of scale, right. Depends on how many sites you have. Yeah. Depends on all sorts of things.

Nic: Yeah. Yeah. That, that, that makes sense. But and the reason why I ask that is, 'cause the next question really is, is it meant to be like a, a SaaS competitor to other, like, well-known household brands like Squarespace or Wix?

Or is it really meant to be a kind of a Drupal to compete with other Drupal products?

Matthew: I would say Aquia sells to enterprise and mid-market customers. And I think that's. Not gonna change. You know, the kinds of people that buy Squarespace and Wix, well, a lot of them are not mid-market and enterprise companies, right?

So it's, it's a different market per interested in a different price point, a different set of features. They're not worried about millions of views in a day and they're not worried about FedRAMP compliance and they're not worried about priority support. You know, all sorts of things that Aquia offers that are really valuable.

For enterprise customers are not offered at low price points by Wix and Squarespace. So I don't think we directly compete with them. I think Wix does have like a Wix enterprise so that they're beginning to try to come upstream and get closer to the same like market segment that we serve. So, you know, we'll see all companies grow and change and, and maybe there will be more overlap with products at some point in the future.

But we're not trying to become a Wix or a Squarespace and I think it wouldn't make sense. That's not what Drupal's good at. You know, I think we, we want to always ask ourselves like, what's Drupal good at? Why would you pick Drupal? How do we accentuate that instead of try to become something else?

John: But I mean, isn't, isn't there a pathway for growth there? Right. So like I always think about it as like. A small business, right? They're, they get started with a Squarespace and they're, they're, you know, building their site. They're doing, you know, maybe they're doing e-commerce, they're doing social media, whatever, whatever they're doing, right?

And then there's some point if they have a successful business where they're like this thing doesn't really work anymore, right?

Nic: Mm-hmm.

John: And like for me, it feels like, it feels like Aquasource could, could, you know, jump into that lower level market, lower level sounds, sounds insulting, but like that small business, that small business market, right?

And be that, that tool that grows with them as and, and advances with them as they grow. So, like, I can understand what you're saying, like you wanna be in the, the mid, mid and up tier, but like. Somebody's gotta be having that conversation somewhere where they're like, well, maybe this is an on-ramp for a lot of new business for us.

Right.

Matthew: You know, I, I'm in those conversations and I'd say right now there's no intention of doing that. Anything can change, but it's a, it's a really different go to market motion. Like you need different sales, marketing support. Like you would require a fairly significant transformation from top to bottom to serve a very different set of use cases and customers.

It's possible. But it, there's a lot of opportunity without doing that.

Catherine: Yeah.

Matthew: So, you know, so I think you gotta kind start there.

Nic: I, I, I guess one question is, and, and maybe, maybe you won't wanna answer this as the VP of innovation at Acquia, but I thinking from a market standpoint, is there a market for whether, it doesn't sound like that's something that Acquia wants to go after, but it does feel like.

Drupal's in a position where,

Matthew: yeah,

Nic: it, there is an opportunity for someone to go after that market again. Sure,

Matthew: sure.

Nic: And

Matthew: I'd love for someone to go after it. And it doesn't have to be Acquia. I, I think that all the time. And I've talked to a number of people about it, like, Hey, why don't you try this out because we need to get Drupal visibility and discoverability kind of up again, you know, people kind of don't discover it organically so much anymore.

And a lot of what we're doing in triple CMS and triple canvas make it more accessible for the non-expert for a low price point for someone who doesn't have a lot of hours to invest in getting it to production ready. So, yes, I mean you can see what, you know, DRE is showing with the marketplace, for instance, in, in previous Dres notes.

But I do think in someone's gotta get some skin of the game, someone's gotta try to like run a business on it for it to move fast enough. When you don't have that incentive. I, I, I have very low confidence in, in success. So maybe we'll have a marketplace. Maybe we'll have some nice themes in there. But humans all run on incentive, you know what I mean?

And it, it's a power capitalism creates powerful incentives. So I would, I would love if someone else went and followed that incentive for that opportunity. I do think it's there.

Nic: Okay. Yeah. That, that it's been on, it's been my mind because that we've been seeing more and more, like, I think it's interesting because I've been seeing, I, I think Drip Yard is a good example of this, right? And

yeah,

the name is escaping me. But I think, was it Drupal Forge, I think is another example. Like you're starting to see some of these businesses popping up that are kind of exploring that market a little bit.

So, you know, I, I, I guess, you know, Aqua, you know, Pantheon platform estate or Upson, you know, I think they have a broader market view sometimes of the hosting level of these kinds of things. And I wonder if they see that market existing, whether they pursue or not.

John: So I think Matthew hit on, hit on a very interesting.

Point there where he is like, ah, aqui is not really looking at this because like there's a big shift in like, you know, employee count business, business objectives, like how, how the business is run, right? I think that probably applies to a lot of other hosting companies as well, right? Like, as we're talking and having this conversation, I'm like, I don't know why, like an investment company isn't like, Hey, we'll give you X number of dollars to like start this company and build, you know, build this, this, this Squarespace.

Squarespace competitor on Drupal, right? Like, I don't know. I don't know. Maybe it's not, maybe there's not a lot of return on investment there in the, in the initial phases, but I don't think we're gonna solve that problem here. But

Matthew: you have to ask, what's the differentiation there? Like, why would a consumer choose to build this off the shelf site on Drupal instead of gonna WordPress or Squarespace or Wix?

What is special about Drupal in that world?

John: Well, I'm gonna, I'm gonna take two seconds and answer that in my opinion. But I don't know that we need to go too much deeper. I mean, I think Drupal is highly customizable and then the ability to export, right? You know, as much as Squarespace says, yeah, you can export your site, but like, eh, can you probably not easily, right?

I think the, the, the real, real like thing to key in it on there is like, Hey, if we were to go from something like aqueous source and then export it to just regular Drupal, right? Like that, that. That's powerful. That's, that's very powerful. Yeah. 'cause like ultimately now you can go and customize and change and do whatever you want.

Right. And grow your business without having to basically replatform your whole, your whole, you know, your whole system.

Matthew: That, yeah. That was part of the theory, but it's also, I don't know how many people that low price point interested in a Wix or Squarespace or WordPress are thinking, I better build something that can be uber customizable later, you know?

Nic: Yeah, true.

John: Yeah, true.

Nic: And, and the truth is building a SaaS platform on Drupal, that is Wix, you're cutting out a lot of that customizability.

Yeah.

For performance reasons, for support reasons.

John: Well, yeah, but I

Nic: mean, I mean,

John: I mean, right. But I think, I think like, you know, I don't know. I, I think aqueous sources is.

It. Got it, got it. Right, right. I think that's, that's where like, okay, you have this, this tool that's using, you know, Drupal Canvas and it's using single directory components and, you know, you guys are absolutely right. Like once, once the marketplace comes online and like there are, there are themes in there that you can buy for this.

Like, it's definitely gonna be very appealing to a lot of people. I'm just, I'm just of the camp. Like, we should also try to shift to bringing in new people that could potentially, you should grow the community.

Catherine: Yeah. Triple C, triple cms, I think does that as, in a way as well as far as getting the, getting the name out there to those type of people.

But I'm interested in, in the advantages that you see for your market, right, Matthew? Like with this SaaS solution, this built on Drupal. I mean, what, what are the advantages of that?

Matthew: So I think. A couple things. One, usually our customers aren't only buying Acquia source, right? And so there's a lot of value in being not just with the same vendor but on the same technology.

Or maybe they have a lot of websites and people who know how to run them on a platform as a service Drupal Stack. And then they have, you know, a couple hundred sites that are simple and they're running them on source, and they like the idea that they can share things between them. And so I, you know, things that aqui are, is good at, like I mentioned some things before, like good at scale, good at security, good at enterprise support.

Another thing I think that we're and usually good at is horizontal scale. That's what Site Factor is good at. We have customers who can spin up 10,000 sites off of one install profile, basically, and manage them in bulk. A unique solution. And we also asked ourselves with source, like, okay, how do we, how do we keep that unique thing, horizontal scale?

Or, you know, you could also call that like governance at scale. And that's actually where some of the tools that are coming out with Dral Canvas and for Drupal Source come from. So we, there's a Canvas CLI tool, for instance. I should talk about the purpose of it, or actually we can talk about the customer problem.

So for instance, imagine you had your 10,000 sites on platform as a service. You have a design system built out a component library, and you want it consistent across all of them. Well, if you're using Canvas on both of them, or even if you're just using single directory components across both of them, we want a way to synchronize between the two.

You know, you can have a dev team with a centralized ci cd pipeline. They roll out these components, they make them available to authors and it doesn't matter what site they're on, whether it's SAS or Pass, they can use it. And then you can imagine, kind of abstracting that to going beyond a component library to a content model, for instance, get syndicated or user roles get syndicated.

So, you know, we're I, I know one, one of the questions, I don't wanna skip ahead John, but you know, it's like where, what's the role of AI in source? And you also like, what's the role of Canvas and Source? So Canvas and Canvas, CLI help us do that syndication across all sites. And this is open source stuff.

You can do this, you don't have to use Acquia source, but you know, there's value to our customers in, in Buying Source and other things because they can bridge that gap and choose what kind of hosting model. They're gonna have, and it doesn't limit them in terms of using the same design system across both of them.

And we're also investing really heavily in the, in the MCP module and in some open source skills because we think that the way that we're gonna all end up interacting with these things in the future is agents. And we, I want it to be a true statement that aqui, or a true statement that Drupal is the best, platform for agents to use or the best content management system for agents to use. You know, it's got the skills, it's got the tools available that, you know, I kind of asked the question redundantly. You answer it, John, like why would someone choose Drupal? Well, I think they would choose Drupal if that were the case.

They would choose Drupal if they knew Drupal's, the best solution I can choose that my agents can choose to build a scalable content management system that's been, that could be dozens or hundreds or thousands of sites that share things between them that I can just trust. So we're trying to make that true and I think there's a, there's a ton of value in that.

Catherine: Do you see like agencies being a good market for Aquasource or do you think it's more, you know, mid-size and enterprise clients that have their own dev teams, whether those dev teams are or internal or outsourced? I

Matthew: think it's really good for agencies. I mean, we're working with a lot of partners on that.

Because, you know, I've worked at, at an agency before, before I worked here, and a lot of, I mean, John or, or Nick, I, I guess all of you can say whether this is true, but I feel like a lot of what the agencies do is they create these kind of internal tools and repeatable processes for delivering for their customers.

And so they've got kind of a layer of IP essentially that they're building on top of Drupal and expertise. And you can still do that with source, you can still have your kind of IP layer of here's how we deliver a content model and a design library and an information architecture and whatever. The other kind of things that you add on top of the framework from a services perspective are, you could still do that, but just on top of source.

But what you can do is you can sell your services at a lower price point. Then you don't have to say, well first take X dollars and set it aside for the cost of running and maintaining Drupal. Right. So it, I think it creates more opportunities for agencies then, because it, it can lower the inherent cost of choosing Drupal in the first place.

John: I, I think that's only one, one factor though, right? I think like, yeah, that makes complete sense. But I also think for, larger businesses and enterprises. Right. Source makes a lot of sense too because, you know, previous client of, of mine basically built Drupal Canvas using layout builder built, you know, a platform similar to Site Factory that kind of did what Aquia Source does now.

Right. And like looking, looking from where we are now, like if Aquia Source were a thing, there are a ton of sites within that organization that could have very easily used a, you know, single directory component component library, like a, a, a slim down version of that and launched and, and, you know, been on the source platform and, you know, lightly maintain, maintained, not necessarily need a bunch of updates and just lived on right in their ecosystem for a lot less cost.

Right. Once you get into that kind of thinking of like, Hey, we have aqueous source. We, it's using our single directory components. Our, it's, it's, you know, using Drupal Canvas for page layout, like it, it checks a lot of boxes. And we can start a site in that, as you said earlier on, and then if we need to push it out to our custom Yeah.

Drupal thing, we can, we can do that. But like I, I definitely think like more, the more businesses that start thinking that way, whether it is an agency or a, or a, you know, larger organization. You know, I think that that, that there's a lot of value there and, and it, it will save, save a lot of, a lot of budgets

Matthew: Yeah.

To gradually something out of it. Yeah. And it's yeah. Yeah. It's, I it's not so hard to get things into it anymore either, but I'm just thinking about something, one of our. Engineers did last week as an experiments with ai and it's related to the content importer and it's kind of related to the import export or portability aspect of this conversation.

Nic: Mm-hmm.

Matthew: Which is, he said he made an agent team, which is that new feature and ran it fully autonomously and, and dangers mode in a sandbox and said here's my, I have a source site. Here are the credentials. Can only interact with JSON API. So, you know, he knows the bounds of like what it can break, it can't delete the site or anything.

And here's a Drupal CMS code base port umami to source and don't come back until you're done. And it's perfect, you know, and it took about 48 hours and it's done, you know, it's pixel perfect and, you know, it just ran and, and you know, it had boundaries of course, to validate its own work. And I started, you know, I had up for a few days.

I was like, what the hell is any of this? So like, you know, what's the value of SaaS? What's the value of pass? Like, oh my gosh, if this is possible. And I had that conversation with him and he was like, no, no, no. I found it really valuable to have SaaS because I, I knew the boundaries that were around the agent.

Like I didn't point it at my infrastructure where it could have just taken the whole thing down. I pointed it at a defined API and I knew the worst it could do was fail to pour umami through taste on API. And I could have the confidence to just let it run free. And then, then I was like, oh, okay. SAS does still make sense in this world.

Like, I, I had to find a new way of thinking about it. But yeah, that's, that's another that's another value where it's the, the aperture, even if you went for platform service, would open more and there's more problems that could grow, even if it's not a human developer doing it.

John: So let's, let's dig in a little bit to the ai, the AI topic.

Obviously we've talked about it quite a bit already, but and I, I don't, I guess I'm not a hundred percent clear on this right now, but aqueous Source, does it, does it have AI built into it? Like, can you interact with a AI assistant to work with your source site?

Matthew: You can. It's there, it's considered experimental.

It's Canvas ai, you know, we built, okay, we built, well, there's a couple ais, but we built Canvas AI as we built Canvas. We actually stopped building it eventually and left it as experimental, partly because I think the architecture of the AI module itself and Drupal. Became a challenge to get high quality consistent results outta Canvas ai.

And also I think because we recognized at the time, like we didn't have the AI expertise to optimize an agent.

John: Sure.

Matthew: And we saw these other tools running so fast, like, like having lovable and having reps that are just really good at site building. And I'm thinking the same thing lately with orchestration.

We're like, okay, we can't race them. How do we stay relevant without competing head to head with them? And that's where we kind of shifted to the strategy of saying we need to go from the outside in. Like what if you could prototype and Figma prototype and Revel prototype and level use 'em for what they're great at.

And then when you're ready you say, put it in Drupal. Put it in source. Could we make that easy? Because then, you know, a customer trusts Rept to make a great prototype. They don't trust it in production. When they have FedRAMP requirements, they're not gonna trust that.

Nic: Right.

Matthew: So, you know, that's where you come over to Drupal and maybe you come over to Acquia.

Then we asked ourselves like, what tools do we need to make that true? And that's when we decided like, wait, we have to expose different information. We need, like this MCP server, we need to think about skills architecture we need to think about, you know, so Canvas, CLI in there and also we open source something called Nebula.

We basically open sourced references for what Drupal canvas components that are well-built look like.

Catherine: Hmm

Matthew: mm-hmm. 'cause we need something to show the agent to say, when you make this in canvas, make it like this.

Nic: Yeah.

Matthew: You know, here's how you organize the props. Here's how you, you know, think about putting it together.

So that's open source. Now we only just, just published it I think like last week. But it, some, some work goes into like building up the resources required for these tools that are advancing really fast to efficiently to. Work with Dral. Mm-hmm. And do a good job with it.

Catherine: One of the things that comes up with my clients quite a lot, and we're dealing with large, large, large, large old sites, you know, sites that have been around a long time, we've amassed a lot of, a lot of pages in them, a lot of data.

They're, they're, they're just behemoth. Right. And one of the things that I think AI could do quite well is help the editing teams of those sites figure out what is in them. Like what did mm-hmm. What is, do we have duplicate content? Do we have we written an article on, you know, yellow cats, you know, how whatever, because when, when we talk about sites that are just massive, massive, massive, and, and have editing teams of, you know, 10 plus, 20 plus people, there's just a complete lack of insight into what is actually sitting on those sites.

It's completely unrelated to what you guys are doing with ai, but I, it's, it's one of the things that I feel like is, is an area that's missed. I, I don't, I don't see that there's any modules out there that are dealing with that. We're doing a lot of that stuff custom. It's, there's a, there's a gap.

Matthew: See, and I don't know if that I'd build a module for it though.

Catherine: Yeah.

Matthew: What, what we should build is the primitives, so that can discover the content easily. We can, I do think it's worthwhile having like a JavaScript embedded chat window or something in Drupal and having a standardized interface

Nic: Yeah. For

Matthew: that. But then you wire it up to you know, open AI or philanthropic as the backend.

Mm-hmm. And I think what'll happen is you, you wire it up to a couple other things, not just LLM as a service, but you'll, you'll wire it up to something like a cowork or a OpenAI frontier mm-hmm. Where it's not just send tokens and get them back from a model. It's more like send it to an agent that itself has other agents behind it.

So there's an orchestration layer behind it and then get a response back. And if you do that, you can kind of rely on the innovation on the outside, but you could say to that agent something complex like what you're suggesting, like, Hey. Just go and find if I have anything on these topics or find generally things where I have redundant topics and it can create multiple agents, hit multiple endpoints, synthesize the data, bring it back, put it in your chat window, which happens to be in Drupal.

And you know, you, your agents weren't built in orchestrated inside of Drupal for that time.

Catherine: Yeah.

John: Mm-hmm.

Catherine: Yeah. Yeah. No, that's true. That's a, that's a really good point. So,

Matthew: so no module in other words.

Catherine: Yeah, yeah, yeah, yeah, yeah. No, that's a great point.

John: Matthew, you've said, said the word orchestration a couple of times, and we had Juergen on a couple weeks ago talking about orchestration in Drupal.

And I'm just wondering what are you more excited about AI or orchestration?

Matthew: That's the same thing for me. Okay. It's, and I, and I wouldn't do it the ECA way either that's another building, something in Drupal that doesn't really belong in Drupal.

John: Mm-hmm.

Matthew: To me, where it's better served by exposing the primitives and but orchestration to me, the way I'm using it is.

Multiple agents working together in an organized fashion to complete a goal.

Catherine: Mm-hmm.

Matthew: And one of the challenges I put out for myself starting last week was like, okay, I'm gonna, I'm gonna change how I'm working with agents. I'm gonna do two things. One is I'm gonna exclusively use teams of agents, not just one, but I'm gonna try to figure out how to use multiple of 'em to the best effect.

And the second is, I am gonna do it in a way so I can, like, you know, we'll kind of one shot it where I'm like, I give you the job and I leave and come back the next day and it's done. I'm not gonna sit there and say, yes, yes, yes. No. Like, I'm, I'm not gonna babysit it. How can I do it well enough and, and actually have this orchestrated layer of agents, like check itself, you know?

Yeah. Have a devil's advocate involved.

John: Just enough autonomy to be dangerous.

Matthew: Yeah.

John: Yeah.

Matthew: And it, I, I, my head has been spinning because of how much. It was I was able to accomplish in the last week. And it honestly like scared me a little bit. And that's when I started having these questions. I'm like, will SaaS still exist in six months or 12 months?

Nic: Yeah, yeah, yeah, yeah.

Matthew: Who was so good at it? To a degree that I never expected. So I'll give you like like two, two or three examples. One is I pointed at a 15-year-old Drupal application. I've been running forever.

Tons of technical debt, right? And I've tried to refactor it so many times and I was like, look at every route in the main menu.

Look at every bundle for every entity type and visit every one until the Google Lighthouse score is a hundred. Like I want you to just don't stop until it's perfectly optimized. And obviously the prompt is a lot longer than that, but I ran it for like three days. It did it and like my, I mean it, it changed custom queries.

It changed pre-processing layers. It changed views, configuration for the prop, proper caching, you know, it rewrote JavaScript what and CSS and fonts and what order they're loaded in and the granularity of it, like all the right stuff. Part of what I said was, since it was the team was create a research team, research Drupal best practices, research front and best practices.

Mm-hmm. The first thing it did is parallelize a ton of research and it read the issue cues, read hundreds of issue cues, and I had hooks in place where, and every time before I implemented a fix, it checked the issue queue. So when looked at the groups module that I used and found the issue and found the patch for it and applied it, but you know, point is I could, I couldn't have done that myself.

I couldn't have hired someone to do that. Like it was incredibly complex. It's something I've tried and failed to do over the course of years, so that shocked me. And I mean, and that was one of like three or four Herculean tasks that I gave it in the last week in this new model. And it succeeded with all of them.

And, and, and you have differently

Nic: before we move on. So running that for three days, what's, what's the cost?

Matthew: 200 a month.

Nic: 200 bucks a month.

Matthew: I, I didn't even exhaust. I, and I had. Three projects running in parallel. Each one with teams of agents. And I have the philanthropic clawed Max, whatever subscription.

No, that's it. I didn't pay for the tokens for all one day. They're gonna, you know, turn the tap off on this. I think this is like a golden age of where they're getting us addicted and then they're gonna pull the rug on, turn it on I

John: to master. Oh, he got, he got AI token into ex non-existence

Matthew: because

John: it cost million dollars,

Matthew: do extra tokens when the new model comes out.

So they gave me a $50 credit also that expires in March. 'cause they, they want you to pull you in. They're like, here's extra tokens. Come use it. Well,

Nic: I guess that's the next question then. How many tokens did that process use? Millions. I mean, rough. Roughly million. How many millions,

Matthew: you know, probably tens of millions.

Catherine: Mm-hmm.

Matthew: On the order of that.

Catherine: I'm playing around with trying to get, I, I wanna pick your brain, Matthew. I'm playing around with trying to get my, my agents to collaborate with one another on like, okay, I, when have I written the right amount of code for me to stop and have, have it code reviewed by this other agent?

And then, so I'm, I'm at a much simpler place than where you're at. I'm just at the beginning stages of having things coordinate and collaborate with one another. And I find it very overwhelming, but also quite exciting because it is doing things incredibly well. I see the, the agent that I use for code review.

For what the other agents are doing, finds things that are just amazing to me. Yeah. Yeah. I mean, I just look at them and go, my, what, why am I here? Like, why, why am I here

Matthew: to guide it? Because it makes mistakes too. And you know, usually you're like, well, it didn't think about it in this way or needed to incorporate this or just take another pass.

But this new perspective, there's still a lot of value coming. Well,

Nic: I, I also, I, I guess I que I guess I question too then Matthew. So it's been three days refactoring a, a very old, it sounds like, highly Legacy app. I assume it changed thousands and thousands of lines of code. Yep. And very deep. Like how do you review something like that?

I don't, do you

just,

Matthew: I don't, I don't review it. I have a different team of agents that writes the test for it, and then I have another team of agents that criticizes the tests to see if it's producing false positives, you know? And so for instance, I did that and it was like, well, I think we need mut mutation testing because we actually need to change the code.

Prove that it fails when we change it, and that proves that the test is actually working when it says passes. And so, yeah, a lot of it is creating competing agents with different incentives. And that, I mean, but think of, it's the same thing you do with a human team. Like here's, I I usually say have a researcher, architect, frontend developer, backend developer, have a quality assurance person and a code reviewer.

The code reviewer is not the same as the QE person. And on top of that, put actual deterministic guardrail. So it literally can't do certain things, even if it somehow finds its way outta the tries to find its way outta the box. But a lot of it is like your creativity with applying the tool. Well, I mean, that's, that's the, this was on, this is only true since December, by the way.

I I don't think it did this, could do this before. Claude 4.5 and 4.6,

Catherine: I don't think it could either. I, again, I, I'm with you a hundred percent my, it shocked me today. There was a moment where I was just shocked.

Matthew: Yes.

Catherine: At what I had done in the last six hours. Yes. Like in the previous six. Not just like what was produced, but in what I personally had done in my role, in my job.

Matthew: Yeah.

Catherine: For the previous six hours. It just looks different. It looks different than anything I had ever thought it would, would be.

Matthew: I, I regularly have existential crisis over this. Yeah, yeah, yeah. Like we, we could, when you get to the next shock and you realize, my God, it's doing something completely different and better than it was a month ago.

What does this mean now and what does that mean for a month from now or a year or two years from now? It's one thing to hear all of these people like breathlessly discussing it on podcasts like this, right? Like you hear about this in the news when you actually put your hands on it and you try to solve a problem that you've spent your life being an expert in and see it solve it so well, so quickly.

It takes more than a moment to recover when, when you see it happen.

Nic: Hmm.

Catherine: I do wish I could pipe my logs into it though. That would be awesome.

Matthew: You can

Nic: Oh

Catherine: yeah, I know. I, I clients you got, you got, you gotta convince clients to let you do it. But

Matthew: I would be single best thing. Sorry. I'll just go AI all the time.

Catherine: Yeah, sorry. Sorry.

Matthew: Century Century great logging tool. Yeah. You can set up you can catch exceptions that there's a module called Raven for Drupal, which just pipes everything to Century.

Catherine: Yeah.

Matthew: They also have something called breadcrumbs where you kind of drop a breadcrumb for, you know, for something that goes across multiple requests so that you can back trace it more easily.

It's more complex. Typically I tell the agents instrument it with breadcrumbs instrument that the H outta everything, I dunno what I can say, instrument everything, but then Century has an MCP server, so then I can just say, there's a bug. Go figure it out. Yeah,

Catherine: I know, I know they're there. We've like, we

Matthew: never,

Catherine: yeah, we use Sentry and all.

Like I know it's there that the issue that I'm running into and, and I think we, we, maybe we have to move on, but we talk about we should, we should talk from my week number four. We can talk all about this, but the issue that I run into is actually just one of governance, right? I mean there, the, the clients for which I would really love to be able to have logs going through an agent refuse to allow that to happen even though technically I could make it happen.

Like they're just not, they're not gonna allow that to happen. And that to me is like such an easy use case for it. As long as you're not logging sensitive data, which we're not, 'cause we're all responsible, right? Then there's no, well maybe then there's, I don't, I find it, it's hard for me to understand why they don't let us just pipe those logs through an agent and, and have that, but they, they don't,

Matthew: can't you just.

Make a copy of the site, copy, run it in a different sandbox environment where you can inter intercept the logs and then run the routes to generate the same logs that you'd get in production and evaluate those.

Catherine: Yeah. But you wouldn't be having like the production data come in, would you?

Matthew: You

Catherine: like,

Matthew: you know,

Catherine: you take No, I can't do that.

Although I am taking notes.

John: All right. So let's, let's circle all the way back here. Go back,

Catherine: let's go

John: to, to Acquia source. And, and Matthew, as we get ready to wrap up here, I'm wondering what you know, we've talked a lot about interesting AI functionality and a whole bunch of, of agents and different different tools and utilities that we can bring into our websites.

I'm wondering you know, looking ahead, what is the future innovation and enhancement for, for aqui source Looks, looks look like. Excuse me.

Matthew: Yeah, I mean, I think it's, it's what we're talking about now. It's, it's making it so it's the best content platform for an agent to operate against. And I think that, you know, that brings the most value and opens up the most use cases.

And I think you'll see nearly everything that we put into source we contribute back to the Drupal community. So you'll be able to see that. First in the Drupal AI initiative and in Canvas and Drupal CMS over the next few months.

John: And when you say for, for an agent, I mean, I'm assuming you mean with a human in the loop, right?

Matthew: Maybe, maybe not.

John: Oh boy. You just made a lot of people real nervous. But that's alright.

Matthew: It's it's the way the world's going. I dunno, you know, it's, don't shoot the messenger.

John: Yeah, but I mean, I, so, okay, so it's the way the world's going. That's, that's fair. Right? But I mean, also I think that like marketers, content creators, site site creators, right?

They want that. They want that ability to approve that, that work. The ability to say like, yes, you did the right thing. Like send it, send it on its way. Do you disagree? Are you seeing things that are different from that opinion?

Matthew: I don. I don't know if I need to agree or disagree because I think our approach is create the tools, create the primitives.

You decide what you do with them. So if you want a human in the loop, put 'em in the loop, put 'em where you want 'em. Depending on what your specialty is, you might make a different judgment call. I don't know where, where or when exactly to do that. For, for marketing. Mm-hmm. Because it's not my expertise.

I have a strong opinion about where to do it for engineering and product. And you, and I think you do need like safeguards that it can't circumvent along with like really good context. So it's very unlikely to try to circumvent and you do need a human in the loop somewhere, but that might be at the very end of the loop, one human one time really quickly.

Like I, mm-hmm. I don't know. There's a lot of qualifiers in there about what exactly that human in the loop looks like,

John: or, or the human could be at the beginning of the loop because you put the right safeguards in place. Yeah. And it's, it's, you know, it's full foolproof, dummy proof. I don't know, whatever, whatever proof you wanna use,

Matthew: but we're the world's still figuring that one out basically.

Yeah. None of us really know that one.

John: Fair enough. Matthew, as always super enlightening conversation. It sounds like you guys are doing super interesting work at Acquia and thanks for joining us.

Matthew: Thanks for having me.

Nic: Do you have questions or feedback? You can reach out to talking Drupal on social media with the handle talking Drupal or by email with [email protected].

You can connect with our hosts and other listeners on the Drupal Slack and the talking Drupal channel.

John: Are you listening right now and thinking, man, I'd really love to be a guest on Talking Drupal? Well, you can do that by going to talking drupal.com and clicking the guest request button in the sidebar.

Nic: You can also promote your Drupal community event on Talking Drupal. You can learn [email protected]. TD promo.

John: Get the Talking Drupal newsletter to learn more about our guest hosts, show news, upcoming shows and much more. Sign up for the newsletter at talking Drupal do com slash newsletter.

Nic: And thank you patrons for supporting talking Drupal.

Your support is greatly appreciated. You can learn more about becoming a patron at talking drupal com and choosing the become a patron button.

John: Alright Matthew, if folks wanted to get ahold of you to talk about AI aqueous source, all the things, work is the best place for them to do that.

Matthew: Slack is where I hang out.

There you go. Slack, mash. That's the place.

John: Cool. Catherine, what about you?

Catherine: Yeah, same Slack.

John: Slack. Way to go. Slack.

Catherine: Yeah.

John: Nick,

Nic: you can find me pretty much everywhere at Nicks van and ICX Va n

John: Sorry, I'm gonna circle back to Catherine. Catherine, what's your name on Slack?

Catherine: It's Bu

John: There you

Nic: go. Which, which one?

Because there's like three or four accounts.

Catherine: The Greek, the Greek one. The one that, the one that's appearing by my video feed here.

John: There you go.

Catherine: Yeah. Yeah.

John: All right. And I'm sorry, Nick, did you, did you go? Yes. Did I? Okay, cool. Well, I'm John Zi and you can find me [email protected], on drupal.org and all the social networks at John Zi.

Or you can find out about [email protected].

Nic: If you've enjoyed listening, we've enjoyed talking. See you guys next week.

John: Thanks a lot everyone.

Catherine: Thank you.