Engineer built data products were never the endgame. Business led data products are.

Shuchi Agrawal's recent piece in Forbes argues that data products are not dead, but that they are no longer the endgame. Value, she suggests, is now created when trusted data is embedded directly into decisions, workflows and AI-enabled action, rather than when the product itself is delivered.

It's a subtle distinction, but an important one. If data products are no longer the destination, what has changed? Have data products reached their limits, or have we simply reached the limits of how we've been building them?

For most of the last decade, treating data as a product represented genuine progress. Instead of leaving data as a by-product of systems and projects, organizations began treating it as something with consumers, owners, quality expectations and a lifecycle. That thinking shaped data mesh, modern governance programs and the product patterns adopted across the major cloud and data platforms.

Shuchi's article doesn't dismiss any of that. It asks a harder question. Once a data product exists, does it actually change a decision? Does it improve an action? Does it give an AI system the context it needs to be useful?

In many enterprises, the honest answer is: eventually.

A business team describes what it needs. That request is handed to a central data team or an implementation partner. It is interpreted, architected, developed and returned for validation four to six months later. Testing reveals that definitions, priorities or context have shifted. Final delivery drifts to eight or twelve months after the original request. The product may be technically sound, but the decision it was meant to inform has already moved on.

Read Shuchi's article carefully and this is really what she is describing. Not a failure of data products as an idea, but a failure of the delivery model that produced them. The concept is not exhausted. The engineer-led factory around it is.

Which leads to the more interesting question. If the destination isn't another engineering deliverable, and it isn't abandoning data products, then what is it?

The next generation of data products is business-led.

A business-led data product starts with a decision, not with a schema. The first question is not which tables to join, but what needs to be decided, what information is required to decide it well and what policies apply along the way.

That inversion matters because the people closest to the decision are the ones who understand its context. They know where definitions are contested, where exceptions live, which numbers are material and what “good enough” actually means for the outcome they are accountable for. Asking them to review a product built for them, months later, is not the same as letting them shape it.

None of this removes engineering. It reframes what engineering does.

Engineering builds the trusted environment: the approved connections, the reusable components, the standards, the policies, the access controls, the guardrails. The business then uses that environment to shape products around real decisions. Governance stops being a review at the end and becomes part of the product from the beginning.

Speed without governance creates risk. Governance without a path to use creates delay. Organizations should not have to choose.

This shift was already needed. AI is making it unavoidable.

AI doesn't understand your business. It understands language.

It doesn't know which customer definition is the correct one. It doesn't know which revenue measure finance has approved. It doesn't know which policy overrides another, or which source is authoritative in a given context. It only knows what the organization teaches it.

Gartner has been direct about the consequences, warning that a lack of semantics leads to inaccurate agents, wasted spending and new governance exposure. McKinsey's work on the foundations for agentic AI points in the same direction: models are not the constraint, the trusted data, workflows and operating model around them are. Deloitte's State of AI in the Enterprise research keeps returning to the same theme — the enterprises that move from experiments to scale are the ones that fix the foundations first.

None of that context arrives by accident. It has to be authored by the people who understand the business, and delivered to AI in a form it can consume.

That is exactly what a data product should be. Not another asset in a catalog, but a governed carrier of business meaning, ready to be used by a person or a machine.

Seen this way, data products are not less important in the age of AI. They are more important. They are the point at which business context, policy and trusted information are made legible to everything that consumes them.

This is exactly the problem Latttice was designed to solve. Rather than asking engineering teams to manually build every business requirement, Latttice enables business teams to create trusted, governed data products around the decisions they need to make, while engineering provides the platform, standards and controls. The result is trusted information created in minutes instead of months, without replacing the organization's existing investments.

The Market Signal

Shuchi Agrawal is right that producing another data product is no longer, on its own, a source of advantage. Advantage now comes from whether trusted information reaches the point of decision in time to matter.

Where we extend the argument is that this is not the end of data products. It is the end of a delivery model that separated the creation of data products from the decisions they were built to support.

Looking back, I think this period will be remembered as the point where the industry stopped asking how to build better data products and started asking who should build them.

Engineer-built data products taught organizations how to manage data more effectively. Business-led data products will teach organizations how to make better decisions. AI will simply accelerate that transition.

Data products were never the destination. They were always meant to become the mechanism through which trusted information reaches the point of decision.