The new year will bring AI adoption in ways that we have not seen before, after a recalibration of what we now know can be achieved within the enterprise. Knowledge graphs that support compound AI will be front and center as they add fuel to converting unstructured information into actionable knowledge. Alongside other tools like GraphRAG that make Generative AI (GenAI) more efficient, they will continue to pave the way for how AI integrates into our daily lives.
Realistic views on what can be done with Generative AI models will bring the year of compound AI
Organizations are beginning to implement the potential of GenAI to solve real problems. In the new year, we will see it adopted in ways not seen before, but when it comes to the adoption of AI for enterprise users, the models are still not sufficient on their own to solve complex problems. Take us humans, for example, we are smarter and more effective with tools, and we have been able to accomplish a lot more with access to calculators, a library, and a computer. We can’t expect language models to do everything we need them to at this stage, especially in an enterprise setting, without the proper tooling. Adding knowledge graphs that support compound AI workloads will allow systems to be broadly leveraged and benefited from within the enterprise.
A revolution of information ranking with GraphRAG
In the early days of the Internet, the primary search engines were AltaVista and Lycos. A search query would index all the words on a page and offer results in a page rank order. Eventually, Google reinvented this by looking at how pages relate to each other. Pages became more important if other important pages were pointed at them. This recursive rule was possible only when you looked at the web as a graph. This is how we ended up with the Google and page rank we know today. Further, when Google started converting textual data into a knowledge graph in 2012, we saw an evolution of how users received structured information about real-world entities when searching.
In the coming year, there will be a similar progression that we saw with the internet from keyword search to search based on network and graph structures. Searches based on converted text to structured representation will also happen with language models, benefiting enterprises hugely. As we progress with GenAI, we’re starting to see something similar with GenAI leveraging RAG, which converts every word or every piece of a document into a vector, allowing us to take a question and map it to the individual words on the document.
I believe the next iteration of the search will move to using a combination of knowledge graph and RAG. What this does is cross-reference documents and quickly find that they have something in common and link it as a connection as it works to respond to a query. Over time, it is likely that most of what we have documented will be converted into structured information that will be put into knowledge graphs that will allow for reasoning to happen when we are asked for a search query. There will be an emphasis on rapidly converting unstructured text information into structured information for symbolic knowledge in order for it to become actionable.
The interface of the internet is changing, our day-to-day life will see AI adoption before the workforce
As someone who grew up on Google, it’s unavoidable to notice that the interface of the internet is starting to shift. The rise of ChatGPT adoption has progressed into becoming the primary mechanism for how the next generation communicates with the internet. As we continue to see this adoption in 2025 and beyond, it will have a significant impact on how industries like advertising evolve to maintain a competitive edge.
As with most innovations of technology, we will implement them in our personal lives first. I believe we will see this happen with personal assistants like Siri or Alexa based on language models that reason and develop natural patterns for our day-to-day habits. As we start to see people rely more on personal assistance outside of work, the expectations of having similar assistants at their jobs will follow suit.
Recalibration of budget for implementing Generative AI in the enterprise
Now that the peak AI hype cycle is behind us, people are much more pragmatic in their approach to GenAI. In the last year and a half, many have spent a large portion of their budgets on GenAI, and they may have put other important areas of the IT footprint and data on the back burner and under-invested. So next year, we will see many organizations calibrating the budget better to do more. Now that we have the visibility and exposure of how GenAI could work or not work for an organization, those businesses can balance out the investment between GenAI and all of the other important initiatives.