Generative AI & Emerging Data Types Impact App Architecture
Back in 2022 I noticed a subtle shift in app architecture that manifested in the market with discussions around new protocols and data. This was not all that surprising, for as I mentioned back then app architectures historically rise and begin to shape the app delivery and security markets every five years, and rise to dominance about five years after that.
In 2023 I thought this shift would make itself evident in about 2025. But I could not foresee the arrival of generative AI, and its explosive effect on the evolution of these architectures.
Which is all to say: that shift is happening faster and is going to have a dramatic effect on, well, everything.
But lest we get all caught up in the architectural evolution, we should really understand what’s driving it: data.
Different kinds of data
Typically, when one says “data” the listener immediately conjures up images of rows of customer and product data stored somewhere in a massive RDBMS. Okay, maybe not all listeners, but lots of listeners do. That’s because the dominant data storage design leans toward structured, tabular data. Open a spreadsheet and you’ll see what I mean. It’s based on rows and columns, and an RDBMS is not really all that different.
We could argue that NoSQL and key-value data stores, object stores, and similar data constructs exist and therefore app architectures do not revolve around tabular data. But come on, they really do. All those other types of data stores are used and used for good reasons, but most of those reasons were ancillary to the application and thus did not significantly alter the overall architectural principles that guided app construction for decades.
But increasing distribution of applications—and hybrid models—drive a lot of needs. Like observability and security, which demand the generation of telemetry data. Distribution requires a way to understand the relationships between distributed data stores, which leads to knowledge graphs. And generative AI demands ways to augment and train models, which leads to the use of embeddings stored in vector databases.
These are four very different kinds of data and come with new ways to store and access it. All four are rapidly becoming “standard components” of a modern application, particularly those that are designed to facilitate the use of generative AI.
- Telemetry is driving significant change on the operations side of the world as we come to understand that columnar formats are a better way to transfer and process real-time streaming data, which is what telemetry data really is.
- Knowledge graphs and graph technology are rising to federate distributed data and understand complex relationships across data sources. This technology brings us protocols like GraphQL that are dramatically changing how we leverage APIs to find and manage data across the distributed enterprise.
- The growing adoption of retrieval augmented generation (RAG) techniques for generative AI are driving vector databases as a part of the application itself, rather than as a data store somewhere deeper in the architecture.
- And finally, let’s not forget generative AI data. An LLM generates new data every time you make a request. That poses significant challenges for security folks, who must figure out how to identify potential threats and problems within dynamically generated unstructured data.
It is data—with its different formats, protocols, uses, and applications—that is driving this change. The result is a headless approach that relies on APIs in ways a lot of folks have never considered before. That is, business capabilities and functions are exposed as APIs without consideration for the traditional presentation layer, which allows a greater range of interfaces to be developed for a more robust set of user devices. APIs make data and capabilities accessible, which is why they are so important to digital transformation.
We see this approach in many modern apps that take advantage of microservices. This is the reason UI frameworks are so prevalent (and cause debate on the Internet). Those frameworks are used to build individual apps that leverage APIs to access functionality and data. They are separate entities from the back-end “applications” they use to drive workflows and deliver value to the market.
All this is driving headless architectures. The shift was already underway, but the arrival of generative AI has illuminated the criticality of data and accelerated it.
This, in turn, will have a profound impact on app delivery and security as new threats and challenges arise that must be addressed to satisfy the needs of business and users.