Beyond Meta AI glasses: World Models, and a Knowledge Graph for Augmented Intelligence
December 2025
by Miguel Santos [Mike] LUPARELLI MATHIEU
Curious about what can be done with the Meta AI glasses? This story is about getting the Meta AI glasses, playing around with the Meta AI SDKs (at high level), and asking questions about what else can be done, and especially, from a product perspective, imagine what will make a difference living in the future. It’s a journey of discovery and product innovation. Let’s go for it!
Obvious
From a developer point of view, it takes a moment playing around with the SDKs (https://wearables.developer.meta.com/docs/develop) to find out that collecting contextual information, such as image and audio, is straightforward.
Thinking about connecting this input to a customizable Generative AI is the obvious next step. Nevertheless, doing the obvious, even better than the incumbent, is not enough for a product to shine among the others.
Not that obvious
So, start asking questions. Why launch AI glasses at such an early stage? Additionally, does having Yann LeCun as Chief Scientist of Meta AI, and being LeCun one of the precursors of “World Models” something to correlate?
So let’s imagine for a moment that the “not that obvious” scenario could be: Meta AI glasses might be a very useful device (one of many) for collecting the kind of contextual information that might be needed for developing “World Models”. Therefore, the “not that obvious” breakthrough might not be related to the IoT by itself, but by the capabilities it can bring. Let's continue with this.
Obvious product
What if a user can augment its capabilities to interact with contextual information? That’s the core purpose of the AI glasses. A triggered Look and Ask (or Take and Look), and natural language interaction with a Generative AI in real time. It is functional, and it is amazing. What’s the problem? From the user perspective, this augmentation is ephemeral. Whatever the user augments the context, that augmentation is lost once the information is no longer actionable (so far).
An obvious product would be to get the augmented capabilities of the AI glasses to collect contextual information. Process that information to have some Generative AI augmented insights, and then store that augmentation in a way that it can be then available to the user for future augmentations. For instance, having an on-device or on-cloud knowledge base so it can then be used by a RAG (Retrieval Augmented Generation).
Not that obvious product
Contextual information means (in this domain) that the user is in contact with some physical environment, and therefore interacting by proximity. Let’s focus on that. Interactions occur by proximity, not remotely. It means that the user can only interact with the context, and with someone else within that context having the same kind of augmented experience with AI glasses.
Whether the user is attending a social event, a conference, a convention, or a sport event, if there is an interaction that wants to be somehow kept and restored in the future (i.e. augmented), that interaction can be processed in real time using the embedded sensors of the AI glasses, then the information stored in a knowledge base (on-device, or on-cloud), and finally, it can be retrieved to augment future engagements.
The IDEATION
A user connecting with another user in the physical world, and both using this technology, will be augmenting their interaction. They both will be assisted by Generative AI in real time, and both will enhance their own knowledge base for future retrieval. In other words, user A will connect with user B, and that connection will contain some information that will be stored for future retrieval, or past information will be retrieved for augment this connection.
Scale this scheme and there will be a network of augmented connections (knowledge graph). Every user represents a node, and each connection has some contextual information stored in their own user’s knowledge base for future retrieval. At the end, there will be a network of knowledge bases. Those connections will be based on proximity, and therefore will be bonded by time and space.
Imagine
Imagine a future where every user is a node, and every node is a knowledge base accessible for a seamless retrieval. Imagine that the AI glasses provide an augmentation of the way that people sense the context, collect information, and retrieve insights, all of this in real time streaming, and completely seamless. Predictive, sometimes. Imagine the power of that knowledge graph, and its implications for building World Models.