Share this article

The emergence of generative AI introduces a new wave of applications that integrate structured and unstructured data, creating a “system of intelligence” layer.

Technology is changing dramatically with the advent of generative AI, which will cause a significant shift in enterprise spending over the next ten years and beyond. Large-scale changes can appear to happen quickly at first, particularly when they create as much of a stir as generative AI has in recent months, but they take time to seep through the layers of the enterprise technology stack.

When businesses put together the components for power and performance, they first invest in the infrastructure layer; judging by the money going into Nvidia and GPU aggregators today, this process is well under way. The focus of development will move to the new products and experiences that will reshape each layer below as adoption (and money) move up the stack.

The application layer transformation is still in its early stages, but early indications point to a significant disruption.

Enterprise apps started to offer more consumer-like experiences long before generative AI emerged by enhancing user interfaces and adding interactive components that would interest regular users and speed up workflow. Applications that were designed as “systems of record,” like Workday and Salesforce, began to give way to “systems of engagement,” like Slack and Notion.

This new breed of enterprise tools, with features like version history, annotation capabilities, multiplayer mode, and metadata, was all about collaboration. These applications also made use of consumer-native viral elements to promote uptake and facilitate easy content sharing both inside and across enterprises. Within these systems of engagement, the core record maintained its inherent value and functioned as a foundation for the increasing amount of data generated at the engagement layer.

We can anticipate even more profound evolution as generative AI shapes the upcoming generation of application products. The initial players bear a striking resemblance to ChatGPT integrators in that they construct simple tools directly on top of generative models, providing brief but instantaneous value. Numerous generative AI products have already surfaced; these have shown tremendous initial growth, but they also have very high churn rates because of their constrained workflows or lack of extra features. These programs usually result in a generative output that is a one-time piece of media or content (i.e., not integrated into a user’s regular workflow), and their usefulness depends on readily available off-the-shelf generative models.

Just starting to take shape, the second wave of generative AI applications will use generative models to integrate the unstructured data found in system-of-engagement applications and the structured data found in system-of-record applications.

Those who develop these products will have a greater chance of building long-lasting businesses than those who enter the market in the first wave, but only if they can figure out how to “own” the layer above the system-of-engagement and system-of-record applications. This is a difficult task given that competitors like Salesforce are already rushing to use generative AI to fortify their underlying layers.

This brings us to the third wave, in which new players develop a defendable “system of intelligence” layer of their own. First, startups will launch new products that use the system-of-record and system-of-engagement capabilities already in place to create value. After establishing a compelling use case, they will develop workflows that, in the end, function as standalone enterprise applications.

The current interactive and database layers won’t necessarily need to be replaced; instead, new structured and unstructured data will be created, which generative models will use to improve the user experience of the product. In essence, this will result in the creation of a new class of “super datasets.”

These products should have integrations that can ingest, clean, and label data as a primary focus. For instance, consuming the knowledge base of already-opened customer support tickets is insufficient to create a new customer support experience. Bug tracking, product documentation, internal team communications, and a host of other features should all be included in a really compelling product. It will be able to extract the pertinent data, label it, and calculate its weight to produce original insights. It will be equipped with a feedback loop that enables it to improve with usage and training both within and between organizations.

When a product does all of this, it becomes very difficult to switch to a competitor because the cleaned and weighted data is extremely valuable and it would take too long to get the same quality with a new product.

At this point, the hierarchy, labels, and weights that go along with the product or model are just as intelligent as the actual product or model. Delivering insights will happen in minutes as opposed to days, with an emphasis on choices and actions rather than merely the synthesis of data. These will be the real generative AI system-of-intelligence products, identifiable by these distinguishing characteristics:

possess a thorough understanding of business processes and the capacity to collect recently generated structured and unstructured data.

Be astute when it comes to using hierarchy, labels, and weights to characterize and process data.

Establish data feedback loops between and among customers to improve their experience with the product.

One important question I frequently pose to clients is, “How does a new product stack up against the other tools you use?” Typically, the most crucial product is the system-of-record, which is followed by the system-of-engagement, and additional tooling at the bottom of the hierarchy.

When money is tight, the least valuable product will be the first to be trimmed, so developing systems of intelligence must offer long-term benefits to be successful. Additionally, they will have fierce competition from market leaders who will incorporate intelligence capabilities enabled by generative AI into their products. To survive, the next generation of system-of-intelligence products will need to combine high-value workflows, teamwork, and the addition of super datasets.

Over the past 12 months, the AI space has undergone rapid transformation, and the industry is picking up new skills quickly. Both closed proprietary models and open-source models are developing at unusually fast rates. It is now the responsibility of founders to create long-lasting system-of-intelligence products atop this quickly changing environment; if done well, the effects on businesses will be remarkable.