Episode 6 of Components Live—recorded in Valencia as part of this 11-part series—features Patrick Bosek and Noz Urbina exploring why graph databases are having a moment—and what that means for structured content and AI.

They begin by contrasting graph databases with traditional relational databases. Where relational systems rely on tables, rows, and predefined schemas, graphs model entities and the relationships between them in a more flexible, self-describing way. The result is a structure that more closely mirrors how humans think—through connections, associations, and context.

To make this tangible, Patrick and Noz use familiar examples: how we recall people through shared connections, the classic “Six Degrees of Kevin Bacon,” and complex product or parts relationships in engineering systems. In each case, graph models excel because they support many-to-many relationships and open-ended exploration—without forcing edge cases into rigid table structures.

This flexibility unlocks something powerful: discovery.

Graphs allow systems to traverse relationships, uncover patterns, and even support inference in ways that are difficult or inefficient with traditional databases. That capability becomes especially important in AI-driven environments, where understanding context and connections is key to delivering accurate answers.

They also highlight an important technical advantage: graph query patterns tend to be closer to natural language. That makes them a strong fit for large language models, which can more reliably generate and execute graph-style queries compared to more rigid SQL-based approaches. The result? More accurate and intuitive question-answering experiences, as supported by emerging research and real-world implementations.

The conversation then zooms out to look at the bigger picture—particularly how Google has evolved its search experience. What was once a list of blue links has increasingly become a dynamic, relationship-driven interface powered by underlying graph structures. It’s a shift that signals a move “beyond the page,” where information is assembled and delivered based on meaning, not just documents.

For technical communicators and content teams, the implication is clear: the future of content isn’t just about managing information—it’s about modeling relationships. And graph-based approaches may be a key part of making content truly AI-ready.

Episode 5 of Components Live—part of the 11-part Valencia series—features Patrick Bosek and Noz Urbina taking on a debate that refuses to die: structured vs. unstructured content for AI.

But instead of picking sides, they challenge the premise.

Patrick and Noz argue that most discussions miss a more important question—what do we actually mean by “structure,” and where does it create value in a system? The answer, as they explore, is highly contextual.

They walk through scenarios where structure clearly improves outcomes. One standout example: tabular data. In large tables, LLMs can struggle to maintain column relationships as they process deeper rows. Without repeated metadata, meaning can break down. In these cases, structure—and even strategically reinforcing it—can significantly improve how AI interprets the data.

On the flip side, they share examples where rigid structure can get in the way. For content like glossaries, converting highly structured, DITA-style entries into clean, natural-language sentences actually produced better results for AI consumption. It’s a reminder that large language models are fundamentally optimized for human language—not strict schemas.

This leads to a more nuanced takeaway: it’s not about choosing structured or unstructured—it’s about knowing when and how to use each.

The conversation reinforces a key principle from earlier episodes: keep content structured upstream. By doing so, teams retain flexibility. Deterministic systems can pull exact values from structured sources (like tables), while transformation layers—using tools like XSLT—make it possible to experiment with different ways of “dosing” content for AI, adapting quickly as models and use cases evolve.

Finally, Patrick and Noz caution against a common trap: storing content only in delivery formats like Markdown, PDF, or HTML. These formats are endpoints, not sources of truth. As channels continue to shift—from websites to AI agents and beyond—structured content ensures that information can be repurposed, recombined, and delivered wherever it’s needed next.

This episode brings the series back to a central theme: structure isn’t about rigidity—it’s about optionality. And in an AI-driven world, that flexibility is what makes content future-ready.

Episode 4 of Components Live—recorded on-site at La Botifarra in Valencia—continues the 11-part series with Patrick Bosek and Noz Urbina diving into one of the most important (and often misunderstood) topics in AI: context.

Rather than treating “structured vs. unstructured content” as a binary debate, Patrick and Noz reframe it as part of a broader challenge—how to effectively navigate and deliver the right context to AI systems at the right time. In this episode, they break down how context operates across different layers of AI, starting with the distinction between training and fine-tuning (post-training) versus how large language models are configured and used at question time.

From there, they unpack one of the most talked-about approaches in enterprise AI today: Retrieval-Augmented Generation (RAG). Instead of expecting an AI model to “know” everything, RAG works by retrieving relevant content from source systems and injecting it into the model’s context window—the limited working memory available during a given interaction.

That limitation is key.

Patrick and Noz explain the concept of the context window as a kind of short-term memory for AI systems—one that can become strained as more information is added, sometimes leading to degraded recall or accuracy, especially in the middle of long inputs. This makes upstream retrieval and selection incredibly important.

To make this more tangible, they compare it to how ChatGPT uses web search: it doesn’t read the entire internet on demand—it retrieves and prioritizes a subset of relevant sources. Enterprise AI systems operate the same way, except instead of the open web, they rely on internal documentation, knowledge bases, and structured content systems.

The takeaway? Success with AI isn’t just about the model—it’s about “context engineering.” By improving how content is retrieved, structured, and delivered into the context window, organizations can significantly increase the factual accuracy of AI-generated responses—often moving from roughly 70% reliability into the high 80s or beyond.

This episode sharpens the focus of the series, showing that the future of AI in technical communication isn’t just about generating content—it’s about designing the systems that ensure AI has the right information to begin with.