What I Learned Building Mautic That Applies to the Agentic Web

In 2014, I started building Mautic. It became the world's first open source marketing automation platform. Acquia acquired it in 2019. Over those five years, I learned things about building infrastructure for a new class of consumer that I did not fully appreciate at the time.

Now I am building Plasmate, an open source headless browser that compiles web pages into structured representations for AI agents. The domain is completely different. The technology is different. The users are different. But the structural problems, the adoption dynamics, and the strategic patterns are remarkably similar.

This is what I learned building Mautic that applies directly to what I am building now.

Lesson 1: Every new consumer class needs its own infrastructure

When Mautic started, marketing automation existed. Marketo, HubSpot, Pardot, and Eloqua all served the enterprise. But they were closed, expensive, and inaccessible to the vast majority of organizations that needed marketing infrastructure.

The insight was not that marketing automation was a new idea. The insight was that a large class of consumers (small and mid-market organizations, developers, agencies, nonprofits) had no infrastructure designed for their constraints. They needed marketing automation that was open, self-hostable, extensible, and free to start with.

The same pattern is playing out with AI agents today. Web browsing infrastructure exists. Chrome, Playwright, Puppeteer, and Selenium all work. But they were designed for humans and human-oriented testing. AI agents are a different consumer class with different constraints: they need structured output (not pixels), token efficiency (not visual fidelity), semantic understanding (not DOM selectors), and speed at scale (not single-session debugging).

The existing tools technically work for agents the same way enterprise marketing platforms technically worked for small businesses. But "technically works" and "designed for" are very different things. When I built Mautic, the opportunity was not inventing marketing automation. It was building marketing automation for the consumer class that existing tools underserved. With Plasmate, the opportunity is the same: building web browsing infrastructure for the consumer class that existing tools underserve.

Lesson 2: Open source is a distribution strategy, not a business model

One of the most important things I learned at Mautic is that open source is how you get adopted, not how you get paid. These are related but distinct.

Mautic grew to millions of installations because it was free and open. Organizations could try it without a sales call, deploy it without a procurement process, and extend it without permission. That distribution would have been impossible as a closed-source product competing against well-funded incumbents.

But the business model was never "sell open source software." It was: build an open core that earns trust and adoption, then offer commercial services (hosting, support, premium features, integrations) to organizations that want operational convenience on top of the open foundation.

I am applying the same structure to Plasmate. The compiler is open source (Apache 2.0). The SOM specification is open. The MCP server, browser extension, LangChain integration, LlamaIndex integration, and all SDKs are open. This is the distribution layer. Every developer who installs Plasmate and every agent framework that integrates SOM expands the ecosystem without a dollar of marketing spend.

The commercial layer sits on top: SOM Cache (a shared semantic CDN for agents), Fleet Orchestration (managed browser infrastructure at scale), and enterprise services. These are operational conveniences that organizations will pay for once the open source foundation has earned their trust.

The lesson from Mautic is that this sequence matters. Open source first, commercial second. Trust first, revenue second. If you reverse the order, you get neither.

Lesson 3: Standards and specifications matter more than implementations

Mautic did not just build software. It contributed to the broader marketing technology ecosystem by establishing patterns that other tools adopted. The API structures, the campaign builder paradigm, the contact lifecycle model, the integration framework. These became conventions that outlasted any single implementation.

I underestimated the importance of this at the time. I thought the code was the product. In retrospect, the specifications and patterns were at least as important. They shaped how an entire category of software was built, including by competitors.

With Plasmate, I am investing in the specification layer from day one. The SOM Spec v1.0 is published as a formal document with a JSON Schema. The Agent Web Protocol (AWP) is a full protocol specification, not just an implementation. The robots.txt extension proposal follows RFC conventions. The .well-known/som.json convention is designed as a web standard, not a Plasmate feature.

This is deliberate. If SOM succeeds only as a Plasmate output format, it has failed. It needs to become a convention that any tool can produce and any agent can consume. That means the specification must be clear, open, and implementation-independent.

I am also participating in the W3C Web Content for Browser and AI Community Group because I learned from Mautic that standards bodies matter even when they move slowly. Getting a seat at the table early means you can shape the conversation rather than react to it.

Lesson 4: The community decides whether you succeed

Mautic's community was the product's greatest asset and its most demanding stakeholder. Contributors built integrations we never planned. Users found use cases we never imagined. And the community's expectations for openness, transparency, and quality pushed us to be better than we would have been alone.

The same dynamics apply to Plasmate, but with an important difference. Mautic's community was primarily marketers and developers. Plasmate's community will be primarily AI agent developers and, eventually, web publishers. These are different audiences with different expectations.

Agent developers care about reliability, speed, and token efficiency. They will adopt SOM if it measurably outperforms the alternatives on the metrics they track (cost per page, latency, task accuracy). The community will form around benchmark results and practical performance, not ideology.

Publishers care about control, cost reduction, and future-proofing. They will adopt SOM-first serving if it reduces their infrastructure load from agent traffic and gives them control over how agents interpret their content. The community on this side will form around economic incentives.

Building for both audiences simultaneously is harder than building for one. But the two-sided nature of the agentic web (agents consume, publishers serve) means that adoption on either side reinforces the other.

Lesson 5: Timing is the variable you control least and matters most

Mautic launched in 2014. Marketing automation had existed since the mid-2000s. We were not early to the category. We were early to the open source version of the category. That timing turned out to be exactly right: the market was mature enough that organizations understood the value of marketing automation, but the open source alternative did not yet exist.

If we had launched in 2008, the market would not have been ready. If we had launched in 2018, someone else would have done it first.

I think about timing constantly with Plasmate. AI agents are browsing the web right now. Cloudflare reports that AI user-action crawling increased by over 15x in 2025 alone. The problem (agents consuming raw HTML) is real and growing. But the solution (structured representations as a web standard) requires adoption by both agent frameworks and web publishers. That adoption takes time.

The question is whether we are building too early (before the ecosystem is ready to adopt) or at the right moment (when the pain is acute enough to drive change). Based on the trajectory of agent traffic and the increasing frustration of both agent developers (token costs) and publishers (crawl load without referral traffic), I believe the timing is right.

But I learned from Mautic that you cannot force timing. You can only be ready when the moment arrives. That means having the specification published, the tools working, the integrations built, and the community seeded before the inflection point. When a major agent framework decides to adopt structured web representations as a default, we need to be the obvious choice.

Lesson 6: Acquisition is not the goal, but you should build as if it might happen

Acquia acquired Mautic in 2019. That outcome was not the goal when we started. But the way we built Mautic (clean architecture, documented APIs, extensible framework, active community, clear licensing) made the acquisition possible and relatively smooth.

I am building Plasmate with the same discipline. Clean separation between the open source compiler and the commercial services. Well-documented specifications. Clear licensing (Apache 2.0 for everything open). A codebase that another organization could adopt, extend, or integrate without depending on us.

This is not because I expect or want an acquisition. It is because building with that level of rigor produces better software. If the code is clean enough for a stranger to understand, it is clean enough for your own team to maintain. If the specifications are clear enough for a competitor to implement, they are clear enough for your community to adopt.

The pattern

Looking back at a decade of building, I see a pattern that repeats:

1. A new consumer class emerges that existing infrastructure underserves.
2. Someone builds purpose-built infrastructure for that consumer class.
3. Open source distribution earns trust and adoption faster than closed alternatives.
4. Specifications and standards outlast implementations.
5. Communities form around measurable performance, not ideology.
6. Timing determines whether you are a pioneer (too early), a leader (right time), or a follower (too late).

With Mautic, the new consumer was the small and mid-market organization that needed marketing automation. With Plasmate, the new consumer is the AI agent that needs structured web content.

The domain is different. The pattern is identical. And the pattern works.

David Hurley is the founder of Plasmate Labs and the creator of Mautic, the world's first open source marketing automation platform. He writes about infrastructure, open source, and the agentic web at dbhurley.com. Research papers are available at dbhurley.com/papers.