Drupal's Dries Buytaert: No Business Model Is Safe From AI

Editor's note: This transcript was edited for clarity and brevity.

The Stable Triangle: How AI Is Disrupting Drupal's Ecosystem

Dom Nicastro: We just came back from DrupalCon Chicago. The Dries Note, 2026 — it wasn't just a state of Drupal. It was a state of the world. You summarized the whole intervention of artificial intelligence into the digital experience world, the open source world, as a combination of incredible excitement and incredible uncertainty and fear. The agency world, developers, marketers wanting to build a website — where do they start? Where does a CMS selection begin now?

Dries Buytaert: The way I framed it in the Dries Note is this mental model I call the stable triangle. For 20 years, the Drupal world has had three things: the product — Drupal, the software; the ecosystem — primarily digital agencies that build websites with Drupal; and the open source community — all the people who build, maintain, and contribute to Drupal. The big question in the Dries Note was: what happens when AI disrupts all three sides of that triangle at the same time?

Dries Buytaert: AI is changing the expectations end users have from their CMS. AI is changing the business model of digital agencies — it's getting harder and harder to charge by the hour. And AI is also impacting how open source communities produce code. It's creating pressure on maintainers because of AI slop. I addressed all three in the keynote. And I was very honest about it — we don't have all the answers.

AI Slop and the Open Source Maintainer Crisis

Dom Nicastro: The best line you had in there was the "can-tribution" versus the contribution. There are a lot of contributions now — I can go into Drupal open source and say, look at this code, I figured it out. But is that a real contribution? Can you actually contribute something valuable?

Dries Buytaert: AI lowers the barrier to contribute. Everybody can fire up an agent, make a change to Drupal, and submit it to the core committers. On paper, that sounds awesome — more people can contribute, which means more contributions, more innovation, more momentum. But while the volume goes up, the quality goes down. These AI agents don't always produce great code.

Dries Buytaert: For many projects that's fine. But Drupal runs millions of websites. We power some of the most mission-critical websites in the world. A relatively small group of maintainers needs to review all these new patches — many of which are now AI-generated. AI can generate a thousand lines of code in no time, and the human needs to go through it line by line, checking for security bugs, scalability issues. When it's AI slop, it's frustrating to maintainers because you're not respecting their time. A key part of my message was: we have to say no to AI slop. People shouldn't submit code they don't understand.

Dries Buytaert: At the same time, AI can be a huge accelerator for people who have expertise. If you know what you're doing, you can use AI to go much faster — and you won't submit slop. It's a double-edged sword. It really disrupts the typical open source model, not just in Drupal, but for all open source.

Drupal's AI Vision: From Marketing Briefs to Background Agents

Dom Nicastro: One of the fascinating things I saw at DrupalCon was agents that monitor a webpage — when it drops to 200 impressions, they flag it. As an editor, I would love that. Someone telling me: that article was great in the beginning, now it sucks, and here's why. You haven't updated it in three years.

Dries Buytaert: That's really where we're going. I showed how you can use AI to build pages, even build components that go on pages — all starting from a marketing brief. You start with a Google Doc that describes what you want, and the AI builds all the pages. One of the things we have in Drupal is what we call the Context Control Center — where a marketer can put their brand guidelines, tone of voice, audience definitions. The agents we ship consult the Context Control Center to build pages, so the quality is higher.

Dries Buytaert: One feature I showed: Drupal AI agents can connect to Google Analytics. You set goals, and if a page isn't performing well, the background agents notice. While you're asleep, they ask: why isn't this performing? They consult Analytics, automatically improve the page, consult the Context Control Center as they make changes, and then present the proposed improvements to a human — because we always want to keep a human in the loop. You'd literally wake up in the morning and see: here are five pages we improved, and here's why. Then you review and accept or not.

Dom Nicastro: Which would be good for me, because I am afraid of GA4. I don't even go there without a cheat sheet.

Dries Buytaert: AI can be democratizing — it can make things more accessible. But the vision is even bigger. Let's say you're a marketer putting on an event and you need 300 people to sign up. You hand the AI agents your event documentation — literally a Google Doc — and say go build me a website. The agents go off with the goal in mind. Their mission is to get 300 people to sign up. They wake up every morning figuring out what they can do better to drive attendance. They create pages, social media content, all of it. That's what we're building toward.

The Agency Question: What Survives When AI Does the Building?

Dom Nicastro: If I'm a CMO and I need to stand up a website in a week, am I going to spend money on an agency? Or am I going to prototype it with AI for $20 a month and set it and forget it? Convince me that agencies still have value.

Dries Buytaert: AI commoditizes the mechanical part of building a website. What it does not commoditize is the expertise and judgment of what a great website is — how you do messaging and branding. Agencies need to evolve their business models away from hands-on keyboard programming work. That's going to go away. Where the value is: understanding customer needs. Often the customer doesn't even know what they want. Agencies have a special skill set to extract that, to bring in expertise from many projects, and help the customer scale.

Dries Buytaert: The other thing I showed in the Dries Note is how you can build a webpage with Lovable in 15 or 30 minutes. It looks good, but it's not robust or scalable. There's hardcoded messaging in the code. What I showed is how — with the help of Drupal — you can move a website built in Lovable to Drupal, also with AI. The agency model becomes interesting again: quickly prototype something for a customer in Lovable, bring that prototype to your sales pitch instead of a slide deck. Then, because those prototypes lack real workflows, permission models, security, scalability, integrations, compliance, and governance — that's where a CMS like Drupal makes the difference. The pitch I gave in the keynote: use AI to prototype fast, but use Drupal to build systems that last.

The New Agency Value Prop: Accountability and AI Configuration

Dom Nicastro: When all's said and done, what's the value prop of an agency? What can they do that we can't?

Dries Buytaert: Agencies have deep expertise in what great looks like. A lot of end users don't know what a secure website is, or a scalable one, or great messaging. They might think they know. They might think AI can get them there. But AI needs to be paired with expertise. The typical Lovable site — you can work on it with maybe three editors. It's built for a single editor. There's no translation. So many things you may need for complex sites. These are the things an agency helps set up.

Dries Buytaert: Something new agencies need to do is configure all the AI capabilities. The AI framework in Drupal has a thousand different settings. You can use different providers and models for different tasks — maybe for automatic translation you use one service, for generating alt text for images you use another, or a local model. How do you optimize your use of tokens? How do you avoid leaking proprietary information to AI providers? Where is a simple model enough versus a state-of-the-art commercial model? Most end users don't know how to do this. That's where agencies make the difference.

Dries Buytaert: I was talking to a customer with a couple hundred websites — multiple product lines under their brand — who wanted to do a brand audit across every page on every site. You can use Drupal AI for that, but you have to set it up. It's a project. Tens of thousands of pages. Every product has brand variants. Somebody has to manage, configure, and run it. You might find hundreds or thousands of changes needed. You can't just let an agent loose. It needs to be directed, overseen. That's the kind of work agencies are going to do. Every agency has to reinvent themselves — rethink their tool chain and processes in this new world.

Dries Buytaert: The other thing is accountability. AI automates a lot of mechanical work, but the accountability stays with humans. Pilots use autopilot flying from Boston to Belgium — almost the entire flight — but we still need pilots because they're ultimately responsible for the passengers. Same with Waze: it gives you instructions, but you're still responsible for not getting into an accident. Agency work is the same. Agents make mistakes. The accountability for security, performance, scalability, conversion rates — that remains human. End users want to pay someone for that accountability. They want to know somebody is looking after all of it.

AI and the DXP Stack: Looser Integrations, Bigger Opportunities

Dom Nicastro: In the DXP world — where AI is making inroads into those integrations — where are the real wins?

Dries Buytaert: We have AI in each individual product — our DAM, for instance, uses AI for all kinds of things. But between products is where it gets interesting. Historically, integrating a CMS with an email marketing tool or a commerce platform required custom development or pre-built configurations. That's changing because AI agents can do a lot of the integration work — through MCP, tool calls, or API calls. It makes integrating platforms much easier.

Dries Buytaert: Think about the event website scenario: an AI agent can add pages to your CMS, launch email campaigns in Marketo, run social media campaigns, log everything in a CRM — all of that evolving into looser integrations using AI agents. It's no longer about connecting APIs and hand-coded integrations. It's about designing responsive, event-driven systems with orchestration tools — things like N8N and other automation platforms. Somebody will have to oversee and coordinate all of that work. That's the agency opportunity.

The Broken Publisher Deal and the Content Discoverability Problem

Dom Nicastro: Content discoverability — AEO, GEO — is on every marketer's mind, and mine too. I'm hiding and running from AI right now. I'm trying to deliver experiences directly: me to the CMSWire audience. Email, video — things AI hasn't fully touched. But I haven't seen great AEO use cases yet. I'm not seeing brands crush it in the answer engines and pull people back to their sites.

Dries Buytaert: It's a problem. I'd call it a broken deal. AI crawlers are extracting content — often illegally, scanning books, all of it — and not returning visitors. The old model was: you publish, Google indexes you, Google sends people back to you. AI chatbots occasionally send someone back, but it's very rare. They're better off summarizing and giving an instant answer. It breaks the economic model. They're extracting content without compensating you. We have to find a new equilibrium.

Dries Buytaert: Right now the deal is pretty broken — you get almost nothing for all the content they extract. One thing that could help: Cloudflare has been writing about a marketplace model, where they'd broker between publishers and AI crawlers. You could say: you can crawl my site, but you need to pay me. The largest sites — Reddit — can negotiate a $60 million deal directly with OpenAI or Anthropic. But you can't do that. You need a middleman. A marketplace, a little like Google Ads. That might provide some relief in the future.

Dries Buytaert: Content behind a sign-up wall — not necessarily a paywall — is another option. More organizations might ask people to register before accessing content, so AI crawlers can't just take it. But it all comes down to what's always been true: make very high quality content. If you have it, people will come and sign up for it. If you have average content that 10 other sites also have, you're in trouble.

Dom Nicastro: That's exactly what I'm learning in the AI era. I call it "interesting content." The stuff that's happening right now — hot news, things AI hasn't consumed yet — those are our top stories. But the foundational SEO content, the "what is customer experience" pieces? Tons of impressions in Search Console. Nobody's clicking. That's what's concerning me heading into 2027.

Dries Buytaert: Traffic is down across the board — it's not news to anyone. If I look across Acquia customers, a lot of them are seeing less traffic. And for most organizations, traffic is a lead gen tool. That said, when a visitor does come from an AI tool, the conversion rate is much higher. Traffic is down, yes — but when AI answer engines do send someone back, it's usually a much better lead. The question is whether that's sufficient compensation. That's the question nobody has answered yet.

Related Article: Brands Are Having a 'Crisis of Faith.' AEO Isn't Making It Easier.

The Tailwind Labs Cautionary Tale: When Your Business Model Doesn't Survive AI

Dom Nicastro: You had a blog post in January — "AI Is a Business Model Stress Test" — tied to the Tailwind Labs story. What happened there?

Dries Buytaert: Tailwind Labs has a developer framework used by many, many websites. They're open source. Their business model was: developers come to their site to read documentation, and they can buy a subscription to access pre-built components. The problem is those pre-built components can now just be generated by AI. The value of what they were selling was completely commoditized. And the second problem: developers read documentation a lot less now — traffic to developer docs dropped around 40% — because if you're writing software with AI agents, the agents already know how to build it. You don't need to consult the docs.

Dries Buytaert: Two problems: less traffic to documentation, and a product that wasn't defensible. Everything that can be specified, AI can build. The value shifts to operating things. Acquia's business model — we don't build websites, we run them. We scale them, secure them, host them. You can ask AI agents to build a website, but they can't actually scale and operate one. Some people just have a business model that doesn't survive AI. Tailwind Labs let go of about 75% of their engineering team. It's too bad — great product, great people. Just a bad business model. That's why I call it a stress test.

Dom Nicastro: And that's really the summation of this whole conversation. AI is ultimately a business model stress test for everyone. It's forced me as an editor to ask: what content really resonates now? What can't AI disrupt? And the answer I keep coming back to is interesting content. Stuff that resonates.