There is a palpable buzz around agentic AI, and for good reason. Systems like Mistral Large Language Model agents and OpenAI's GPT-based assistants are pushing the boundaries of interoperable and purpose-built LLMs. At Living Assets, we find agents particularly exciting when it comes to increasing and sharing context windows to help Agents learn, adapt, and perform tasks autonomously. An AI agent continually updating itself with new knowledge is a pleasant reality. It holds so much potential to transform how we interact with the web. However, to fully unlock AI agents as more than glorified RAG systems, we need to address some critical gaps in system design and the scope of what these agents can access.

The Offline Knowledge Gap

Agentic AI promises A LOT, but its promises unknowingly rely on the flawed assumption that the most valuable human wisdom is available in text, either publicly online or in corporate knowledge repositories. In reality, the vast majority of knowledge remains offline, locked away in metadata --from ebooks, digitized archival footage, photographs, niche research papers, state-funded documents, artists' archives, and file storage repositories.

Analogy: Imagine you're in a massive library that holds every book ever written. But instead of accessing all the shelves, you can only browse the front desk, where a limited selection of popular titles is displayed. That's how today's AI agents operate—they can only reach what's readily available on the surface web in text format. Even innovations in the AI space tend to focus on expanding the effectiveness of text-based knowledge acquisition for AI systems. Too bad that only a tiny fraction of all human knowledge exists in online text and the amount of information generated by humans in other formats is growing each day.

The Role of Knowledge Graphs

At Living Assets, we enhance AI Search agents to become genuinely intelligent and versatile. In support of this vision, we help agents access a far richer and more interconnected web of information than they currently do. And the most valuable information for AI Agents comes from creative humans!

Today's AI agents often operate in silos and rely on fragmented and isolated datasets, which restricts their ability to understand complex relationships and make informed discoveries on behalf of their human researchers based on a holistic view of information.

Inter-Agent Knowledge Graphs

Recent advancements in AI knowledge graphs have further highlighted the need for more comprehensive knowledge sources. For instance, a research team at IBM published their way of improving "slot filling" to build more interconnected AI-driven knowledge graphs. But again, if AI agents can only access their knowledge graphs, they're still limited.

At Living Assets, we create Agentic knowledge graphs that can be shared, monetized, and integrated between discrete agents. Enabling inter-agent sharing enhances each Agent's understanding and capabilities. This approach isn't just about making individual agents brighter; it's about fostering a collaborative network of agents.

If you're entirely new to Search Agent Optimization—read our article.