Co-Creative Intelligence: When AI Agents Help Build Their Own World
There’s a lot of attention on AI agents right now. Traditionally, we’ve thought of them in simple terms: the agent is the dynamic, intelligent part; the environment or game is the static space in which the agent operates.
But with agents that work with knowledge graphs and ontologies, that classical view breaks down. Here, the relationship is not one of action in a fixed environment, but of co-creation. The agent helps build the environment itself: it forges the network of connections, distils ontological abstractions, and prunes inconsistencies. Once that structure exists, the agent navigates it, leveraging the very framework it helped create. Ultimately, these two activities become deeply interwoven: navigating the network is itself part of building it.
In this sense, the environment - the network - is not a static, passive backdrop. It is dynamic, evolving in lockstep with the agent’s reasoning, knowledge and memory. The agent is simultaneously creator and explorer, both "architect and inhabitant" of its own world.
This dynamic mirrors what neuroscientists and theoretical biologists describe in frameworks like Active Inference. The Free Energy Principle tells us that intelligent systems, including humans, don’t simply receive data from their environment. We actively sample the world, make predictions about our sensory inputs, and act to fulfil those predictions. Our beliefs shape our actions, and our actions reshape our reality.
The human element is especially critical when it comes to the conceptual heart of the knowledge graph: the ontology. Ontological modelling is not just a technical task but a strategic act of commitment - a deliberate stance on how to see the world. It’s about deciding what matters most in the context of your organisation. Do you model customers by transactions or by engagement? Is a “project” defined by its budget, its team, or its deliverables?
These are not trivial choices. They are expressions of a unique worldview. Just as you can slice an orange in different ways - each valid, each revealing a different structure - so too can you model your data in ways that reflect your priorities and strategy. These are decisions in which we want humans deeply involved.
The real source of competitive advantage in the coming years won’t be a slightly better algorithm. It will be having a better model of the world. By making clear, thoughtful ontological commitments, organisations can differentiate themselves from the undifferentiated, “gestalt” intelligence of giant foundational models.
Ultimately, there is a certain magic in this process of co-creation, where the world becomes what it is through our interaction with it. The most successful organisations will be the ones that don’t just use AI to operate in the world, but partner with AI to help shape the world.
