Business understanding is becoming enterprise infrastructure.

The Market Signal in Brief

In late July 2026, Precisely published a piece arguing that most enterprise data estates were designed for a consumer that no longer holds the final say. Analysts, dashboards and reports all assumed a person on the receiving end, someone who could pause when a number looked wrong, ask a colleague, or check a second source. Agentic systems remove that pause. The standard shifts from data that is good enough for a human to interpret to data that is good enough for an autonomous system to act on without review.

The article is short, but the argument underneath it is consequential, so it is worth breaking down properly before offering a view on it.

I believe this article signals something much larger than Agentic AI. It signals the beginning of a new enterprise capability: business understanding itself becoming infrastructure. That is the shift I believe enterprise leaders should be paying attention to, because it changes the conversation from preparing data for AI to operationalizing the knowledge that already exists across the business.

What the Precisely Article Argues

Precisely makes four moves. Each one is worth isolating, because each carries a different implication for how enterprises should respond.

One: the consumer of enterprise data has changed. Data estates were optimized for interpretation, not execution. A person reading a report performs a large amount of silent correction that was never specified anywhere. An agent performs none of it. Remove the interpreter and every assumption that was carried informally has to be made explicit or it simply disappears from the process.

Two: context is the first gap. An agent needs to understand what data means in relation to the business, not merely what a field contains. Whether a customer is active, whether revenue is gross or net, whether an exception rule applies in one market and not another. These are business definitions, and in most enterprises they exist in several versions at once.

Three: freshness is the second gap. A decision made on a two week old view is not autonomy. The article is right to raise it, and the practical difficulty is that freshness is not a single enterprise standard. It is a property of the decision being made, which means the business has to set the threshold rather than inherit whatever the pipeline happens to deliver.

Four: traceability is the third gap. When an agent acts, you have to be able to reconstruct what it saw, what it reasoned from and why it did what it did. This is a stricter requirement than conventional lineage, which answers where data came from rather than what was visible at the moment of action, under which policy and on whose authority.

The article then makes an observation that is easy to read past and is arguably its most important line: most of the context an agent needs is not missing. It already exists inside the business as tribal knowledge that has never been formalized into something a machine can consume.

That observation reframes agentic readiness as a question about the business itself rather than a question about the data platform, and it aligns closely with what we are seeing across enterprises in Asia Pacific and Japan.

Harry's Perspective

I want to start by acknowledging Precisely for framing this well. There is no shortage of commentary on agentic AI. There is far less commentary on the unglamorous question of what an agent actually needs from the enterprise before it can be trusted to act. This article asks that question directly, and it answers it without reaching for a product pitch.

Where I would extend the argument is on where the work has to happen. Read quickly, the three gaps sound like a data engineering backlog: enrich the context, shorten the refresh cycle, improve the lineage. Read carefully, they are not engineering problems at all. Context is a business definition question. Freshness is a decision-consequence question, because the acceptable latency for a fraud decision and a category review are not remotely the same, and only the business can set that threshold. Traceability is a governance and accountability question. Engineering can build the mechanism for all three, but it cannot supply the answers.

None of this diminishes the importance of data enrichment, metadata, lineage or governance. In fact, they become even more important as organizations move toward Agentic AI. The question is not whether enterprises need these capabilities. They clearly do. The question is how they are brought together into something the business can own, trust and reuse consistently.

In my experience, this happens through trusted, fit-for-purpose business-led data products, where data enrichment, metadata, governance, lineage and business context are not managed as separate disciplines but are operationalized together around the decisions the business is trying to make.

A trusted business-led data product is not simply a curated dataset. It combines business meaning, data enrichment, metadata, governance, lineage, policies and decision context into a reusable enterprise asset that can be consumed consistently by people, analytics, operational applications and AI.

This is the distinction I believe is becoming increasingly important. The market is correctly investing in stronger governance, richer metadata, better lineage and greater business context. What organizations are now asking is how those capabilities become operational. Business-led data products provide that operational model.

That distinction matters because it changes who has to be in the room. In most of the programs I see across Asia Pacific and Japan, agentic readiness is still being run as a platform workstream with business stakeholders consulted at review points. That sequencing is backwards. The business understanding is the scarce input. The platform is the easier half.

It also changes how progress should be measured. Counting onboarded sources or documented tables tells you very little about whether an agent can be trusted with a decision. The useful measure is narrower and harder: for a specific decision, is there a named business owner, an agreed definition, an enforceable policy, a stated freshness requirement and a record that can reconstruct what the agent saw. Very few organizations can answer yes for even one decision, which is a more honest readiness score than most maturity assessments produce.

This is also why I read the Precisely piece as a signal about something larger than agentic AI. Across Gartner, McKinsey, Microsoft, Databricks and Snowflake, the same pattern is surfacing from different directions: the conversation is moving past data quality, metadata and governance frameworks toward how organizations operationalize their own business understanding. Across the organizations I work with throughout Asia Pacific and Japan, that is where I believe the next competitive advantage will be built.

The quiet assumption behind every enterprise data estate

For thirty years, enterprise data architecture has carried an unstated assumption: a person would be there at the end. That assumption shaped everything. It is why a column named CUST_STAT_CD was acceptable, why three systems could hold three different definitions of an active customer, and why a quarterly reconciliation meeting was considered a control rather than a symptom. Human interpretation was the last mile of the data supply chain, and it was free.

It was never actually free, of course. It was simply unmeasured. The cost showed up as delay, as rework, as the fortnight an analyst spent proving why two reports disagreed. In a quarterly business rhythm, that cost was tolerable. In a business rhythm measured in seconds, it is not.

Agentic systems make this visible in a way that dashboards never did. An agent does not know that the finance team stopped trusting a particular field two years ago. It does not know that a region reports revenue gross while another reports it net. It does not know that a customer flagged as active in the CRM may have been in collections for ninety days. It executes against what it can see, at speed, across systems, and it does so without hesitation.

Why models stopped being the differentiator

Three years ago, model choice was one of the defining strategic decisions for enterprise AI.

Today, as the leading models rapidly converge in capability and organizations increasingly use multiple models for different workloads, the source of competitive advantage is shifting elsewhere.

Anything that can be replaced or upgraded as the market evolves is unlikely to provide a lasting strategic advantage.

What cannot be replaced is an organization's understanding of itself.

No vendor can supply your definition of a profitable customer, your interpretation of a regulatory obligation in a specific jurisdiction, or the sequence of events that constitutes a churn risk in your industry. That understanding is proprietary by definition, and it is the one input to an AI system that competitors cannot buy.

Gartner has been direct about the consequence of leaving business semantics unformalized, warning that the absence of shared semantics leads to inaccurate AI agents and unnecessary spending.

McKinsey reaches a complementary conclusion from a different perspective, arguing that organizations need reusable, governed data products, semantic foundations and AI-ready data if Agentic AI is to scale beyond isolated pilots.

Together, they point toward the same destination: enterprise AI depends less on increasingly capable models than on increasingly trusted business understanding.

From tribal knowledge to operating infrastructure

The Precisely article makes a point that deserves more attention than it usually receives: most of the context an agent needs already exists. It is not missing. It is unformalized. It lives in the head of the person who has run the pricing model for eleven years, in the exception rules a regional controller applies at quarter end, in the glossary a governance team maintained for two years before the program lost funding.

Calling that tribal knowledge understates its value. It is the accumulated understanding of how the business actually works, and it is usually accurate. The problem is that it is stored in a format only humans can execute, and it degrades every time someone leaves.

Treating business understanding as infrastructure means changing its status inside the organization. Infrastructure is specified, owned, versioned, monitored and funded. It has a service level. Someone is accountable when it fails. Applied to business understanding, that means definitions carry an owner and a version history, policies are expressed once and enforced everywhere they apply, lineage is a runtime property rather than a diagram, and trust is a signal that travels with the information instead of a reputation held informally by a team.

The test is simple and uncomfortable. If the three people who understand a critical decision left the organization tomorrow, would the decision still be made correctly the day after? If the answer depends on those individuals, the understanding is not infrastructure. It is a dependency.

The consumer has changed, and so has the standard

A human reader of a report performs an enormous amount of silent correction. They notice that a figure looks unusual. They recall that a system was down last Tuesday. They apply judgment that was never written down anywhere. Remove them and every one of those corrections has to be encoded somewhere else, or it simply does not happen.

This is why agentic readiness is not a data quality initiative with a new name. Data quality asks whether a value is correct. Business understanding asks whether the value means what the consumer thinks it means, whether the consumer is permitted to use it for this purpose, and whether the resulting action can be explained afterward. Those are governance and semantic questions, and they are answered in the business, not in the pipeline.

Traceability deserves particular emphasis. Conventional lineage answers where data came from. Agentic environments require something stricter: a reconstruction of the decision. What information was visible at the moment of action, under which policy, at what level of trust, and on whose authority. In regulated industries across our region, that requirement is arriving faster than most programs are prepared for.

Enterprise AI success is not determined by the intelligence of the agent. It is determined by the quality, governance and business context of the information the agent can access.

A different question for the executive team

Most AI strategy conversations still begin with capability. Which models, which platform, which use cases. Those questions are answerable and comfortable, which is part of their appeal. The harder question is this one: where does our business understanding currently live, who owns it, and what happens when a machine needs it at three in the morning without anyone available to interpret it?

Organizations that can answer that question are building something durable. Every decision they formalize makes the next one faster. Every definition they govern reduces ambiguity across analytics, reporting and automation simultaneously. The advantage accumulates, and unlike model capability, it does not reset with the next release cycle.

Organizations that cannot answer it will continue to run impressive pilots that never reach production, because the sophistication of the agent was never the constraint. The constraint was that the business had never written down what it knows in a form anything other than a person could use.

This is precisely the problem Latttice was built to solve. Rather than asking engineering teams to reverse engineer meaning from systems, Latttice lets business teams operationalize what they know as trusted, governed, reusable business-led data products, while engineering provides the platform, standards and controls underneath.

Business Understanding Needs an Operating Model

Long before Agentic AI became the industry's dominant conversation, our focus at Data Tiles was never on building another data platform. We believed organizations needed an operating model that allowed business understanding to become a reusable enterprise capability. Agentic AI has not changed that belief. It has simply accelerated its importance.

The conversations I have with enterprise leaders across Asia Pacific and Japan reinforce this almost every week. The challenge is rarely a lack of AI capability, and it is almost never a lack of enterprise data. The challenge is turning business knowledge into something that becomes operational across the organization.

Business understanding still lives inside documentation, spreadsheets, governance committees and the heads of a relatively small number of experts. None of those are places an enterprise can reliably consume knowledge from at scale, whether the consumer is a person, an analytics platform or an AI agent.

In conversations with CIOs, Chief Data Officers and business leaders across the region, I rarely hear concerns about whether AI models are improving. I hear concerns about whether their organizations can consistently provide those models with trusted business meaning across regions, business units and regulatory environments.

Increasingly, I am seeing organizations realize that the practical way to operationalize business understanding is through trusted, business-led data products rather than through additional documentation or governance processes. Business understanding does not become operational through documents. It becomes operational through trusted, fit-for-purpose business-led data products. Business-led data products are the reusable enterprise mechanism through which trusted business understanding becomes available consistently across people, analytics, operational applications and AI. That belief has shaped the way we have built Data Tiles from the beginning.

Latttice is the Data Product Workbench where business teams operationalize what they know as trusted, governed, reusable business-led data products. Engineering continues to provide the enterprise platforms, integrations, security and infrastructure. Business teams provide the definitions, governance, policies, ownership and decision context that only they can provide. Those business-led data products then become reusable enterprise assets that support reporting, analytics, operational applications and AI from exactly the same trusted foundation.

Lenz extends that foundation into AI.

Rather than asking every AI project to rediscover business meaning independently, Lenz enables AI agents to consume the same trusted, governed business-led data products that people already use.

Rather than every AI initiative creating its own interpretation of the business, Lenz enables AI capabilities to inherit an already trusted business understanding.

That consistency is what allows organizations to scale AI without continually rebuilding business understanding.

Together, Latttice and Lenz provide what I am increasingly seeing enterprise leaders across APJ ask for: an enterprise AI operating model. One where business understanding becomes a reusable enterprise capability rather than something recreated project by project.

This is what Cameron Price is describing when he talks about the transition from intelligent agents to intelligent organizations. An intelligent organization is not defined by how many agents it has deployed. It is defined by whether every person and every AI capability is working from the same trusted understanding of the business.

It is also why we talk about the Decision-Driven Enterprise rather than AI readiness. Readiness is a checkpoint. The objective is enabling every important business decision with trusted, governed, fit-for-purpose business-led data products that people and AI can consume consistently. An organization that gets that right can evolve its AI over time, adopting new models and new agents as they arrive, without rebuilding its understanding of itself each time.

The Market Signal

Precisely is right that agentic readiness is a different standard, not a higher one. Where we extend the argument is in where the work has to happen. It is not primarily a data engineering task. It is the formalization of business understanding into something that can be owned, governed and served continuously.

Looking across the wider market, Gartner is talking about semantics. McKinsey is discussing AI-ready data products. Microsoft continues to emphasize trusted enterprise knowledge, semantic understanding and reducing fragmentation as foundational capabilities for enterprise AI. Databricks has introduced Business Semantics. Snowflake continues to emphasize governed enterprise context for AI.

These are not isolated developments. Together they point toward the same destination. Enterprise AI is becoming less about teaching machines how to think and more about helping organizations operationalize what they already know.

Looking back, I suspect we will not remember this period as the moment AI became intelligent. We will remember it as the moment organizations realized their own business understanding had become strategic infrastructure.

The next competitive advantage is not a better model. It is an organization that has made its own understanding operational.

Organizations that operationalize that understanding through trusted, governed, fit-for-purpose business-led data products will not simply deploy more AI. They will build organizations capable of making better decisions consistently, transparently and at scale.

That is ultimately what we mean when we talk about becoming a Decision-Driven Enterprise.

Technology does not create business understanding.

It operationalizes it.

Organizations that recognize that distinction will build AI capabilities that continue to evolve without continually rebuilding the trusted business understanding that underpins every important decision.