From Trend to Possible: What Gartner's 2026 Data & Analytics Trends Look Like in Practice

Executive Summary

Across its 2026 research, Gartner describes a market moving toward AI agents, decision governance, operational AI governance, agentic data management, real-time data, semantics and contextual AI. Read individually, these look like separate trends. Read together, they describe a single architectural question that most enterprises have not yet answered: how does trusted organizational understanding reach the point of decision with its business context and governance still intact?

I want to be careful about how I respond to research like this. It is tempting for a vendor to map each trend to a product feature and declare victory. That is not a useful contribution, and it is not what I think is interesting here. Gartner is describing where the market is heading. What I find particularly interesting about the 2026 trends is that several of the capabilities now becoming important to enterprise AI are things we have already been working to make practical.

That is a narrower claim, and a more honest one. We have not solved enterprise AI. Nobody has. But in a few specific areas, the discussion has moved from what will be required to what is already possible, and it is worth being precise about which areas those are.

Why This Is a Second Signal, Not a Repeat of the First

In my earlier Market Signal on Gartner's 2026 Data & Analytics trends, I wrote about what I saw as the AI readiness gap underneath them. Agents, semantics and platform convergence were moving quickly, but many organizations still lacked the trusted business understanding required to put those capabilities to work with confidence.

Gartner's June 16 articulation of the six trends makes the next part of that story considerably more concrete. We are now talking explicitly about decision governance, operational AI governance, agentic data management, real-time intelligence and context-rich retrieval. For me, that moves the conversation forward. The question is no longer simply what enterprise AI will require. In some areas, we can now ask what those requirements look like in practice.

Before getting anywhere near what we build, the six trends deserve to be read on their own terms.

Sovereign AI: Control Is Becoming Part of AI Architecture

Gartner's first trend, Sovereign AI Accelerates, reflects the growing expectation that organizations and nations can exercise control over where AI and the data behind it operate, under whose jurisdiction, and on whose infrastructure. This has moved from a policy conversation into an architecture conversation, because AI is no longer confined to a lab environment. It is being embedded into operations.

I am not going to claim we have a sovereign AI answer, because that would be dishonest and the topic is much larger than any single product. What I would say is that the questions leaders are being asked have changed in character. Where is enterprise data actually being processed? Which models and providers are involved? Which jurisdictions apply? Who controls access? What organizational knowledge is leaving the enterprise boundary when a prompt is sent somewhere? What dependencies are being created that will be difficult to unwind in three years?

None of those are questions about model capability. They are questions about control. AI architecture is increasingly becoming an architecture of control as much as an architecture of capability, and I suspect that shift will outlast most of the current tooling debate.

Decision Governance: The Market Conversation Has Moved Toward the Decision

Gartner's second trend, Reducing AI Agent Risk with Decision Governance, is the one I have spent the most time thinking about. Gartner describes AI agents increasingly executing strategic, tactical and operational decisions, and predicts that by 2029 explicitly modeled business decisions will be five times more trusted and 80 percent faster than ungoverned decisions.

I want to resist the temptation to read that as validation of anything we have said. It is more interesting than that. The market conversation itself is moving closer to the decision. For the better part of two decades, the governance discussion has been a discussion about governing data. Who owns it, how is it classified, how is it protected, how is quality measured. All of that remains necessary. But if agents are going to participate in decisions, an organization also has to be able to answer a different set of questions. What decision is being made. What information supports it. What business context is required to interpret that information correctly. What outcome is intended. What authority does the agent actually have. What evidence needs to remain afterward. And what governance applies to the decision itself, rather than only to the data feeding it.

This is why we keep telling customers to start with the decision and work backward. That principle does not change based on who or what makes the final call. It applies whether the decision is human-made, AI-assisted, agent-recommended or agent-executed. It is also the argument behind our move from Data-Driven to Decision-Driven, which I increasingly think is going to be an organizing idea for the next few years.

AI Governance Platforms: From Documenting Governance to Operating It

Gartner's third trend, Driving Trust with AI Governance Platforms, argues that traditional assurance approaches are under real pressure as AI regulation, emerging risks and autonomous agents increase complexity, and that organizations need to operationalize governance rather than describe it.

That phrase, operationalize governance, is the important one. There is a difference between documenting governance and executing governance, and AI is exposing it. Catalogs, policies, ownership models, classification, stewardship, quality frameworks and lineage all retain their value. The weakness is not in the artifacts. The weakness is in an environment where those controls only take effect if a human being remembers them, finds them, and interprets them correctly at the moment data is used. That limitation was already visible when consumption happened at human speed. Systems operating at machine speed make it impossible to ignore.

Gartner's March 2026 predictions research points in a similar direction, forecasting that by 2030 half of organizations will use autonomous AI agents to interpret governance policies and technical standards into machine-verifiable data contracts. Read plainly, that is the market saying governance has to become machine-operable. At the Orlando summit earlier this year Gartner put it more bluntly still, describing data governance as the single point of failure for organizations' AI ambitions.

This is the ground our Active Governance series has been working over: governance that is present at the exact moment data is accessed, combined, shared, or consumed, not before and not after. I will come back to what that looks like in practice, but the conceptual shift matters on its own. Governance is moving from documentation toward execution.

One caution worth adding. Gartner also predicts that by 2027, 40 percent of enterprises will demote or decommission autonomous AI agents because governance gaps are discovered after production incidents, and warns that applying uniform governance across all agents is itself a route to failure. Appropriate governance does not mean identical governance. It has to reflect context, autonomy, risk and intended use.

Agentic Data Streaming: Faster Is Not the Same as Better

Gartner's fourth trend, Agentic Data Streaming Powers Real-Time Intelligence, expects adoption of data streaming for agentic AI to rise from under 15 percent in 2025 to more than 60 percent by 2028. I am not a streaming architecture commentator and I will not pretend otherwise. What interests me is what that acceleration does to decision-making.

Real-time data without real-time context and governance can simply allow an organization to make the wrong decision faster. Speed is not the outcome. Better decisions are the outcome. I have seen organizations push latency down without pushing meaning and control along with it, only to end up with a faster argument about whose number was right.

The faster data moves into decisions, the more important it becomes that governance moves with it.

Agentic Data Management: Machines Are Becoming Participants in the Data Estate

The fifth trend, Streamlining Operations with Agentic Data Management, describes AI agents taking a more active role inside data management itself, acting in real time, detecting patterns and making recommendations, with Gartner emphasizing that strong governance and continuous monitoring remain essential.

The implication runs deeper than automation. For decades, humans built, moved, curated, interpreted, governed and consumed enterprise data. Machines are now beginning to participate in those same activities, which raises a question the industry has not really settled. What happens when machines increasingly participate in managing the data that other machines consume?

The answer cannot simply be more automation. If human intervention is reduced, then trust, provenance, business context and governance become more important rather than less, because there are fewer moments where somebody notices that something looks wrong.

GraphRAG: The Deeper Signal Is Context

Gartner's sixth trend, Handling Complex Use Cases with GraphRAG, addresses the limits of traditional retrieval augmented generation when questions are complex and context-rich. Combining knowledge graphs with large language models improves the retrieval of relationships and context, and Gartner predicts 40 percent of enterprises will have leveraged GraphRAG techniques by 2029.

I would not turn this into a technical article about graph retrieval. The deeper signal is context. Gartner made the same point from another angle in May, observing that agentic AI outcomes depend on context including semantic representations of data, and predicting that organizations prioritizing semantics in AI-ready data can increase agentic AI accuracy by up to 80 percent and reduce costs by up to 60 percent by 2027.

Enterprise understanding is a bigger category than semantics alone. It includes business meaning, relationships, definitions, rules, ownership, permissions, assumptions, governance, lineage and purpose. A model cannot reliably reconstruct all of that from raw data, no matter how good the retrieval technique becomes, because most of it was never written down in the data in the first place.

AI does not simply need more enterprise data. It needs more enterprise understanding.

What Connects the Six

Read separately, these are six different conversations for six different teams. Read together, they are variations on one question. Sovereign AI asks where AI and data can operate and under whose control. Decision governance asks what decisions AI can participate in and under what authority. AI governance platforms ask how controls become operational. Agentic data streaming asks how information reaches AI at the speed at which it acts. Agentic data management asks what happens when machines participate in managing enterprise data. GraphRAG asks how AI obtains the relationships and context required to understand enterprise information at all.

The common denominator isn't simply AI. It is trusted organizational understanding at the point of decision.

In the earlier Market Signal, I argued that the real gap was not access to AI capability. It was whether organizations had the trusted data, business meaning, governance and ownership required to use that capability confidently. Gartner's more detailed trends make the next question clearer. How do those foundations become operational at the point where people and machines actually make decisions?

Where the Trend Becomes Practical

Latttice enables organizations to create trusted, governed, reusable data products around real business requirements. Those products bring enterprise data together with the business context, logic and governance required to use that information appropriately. The product is not a dataset with a label attached. It is the information, the business meaning, the context, the rules and the purpose that determine how it can be used.

It is worth being clear about what this does not require. Your data platforms, warehouses, lakehouses, catalogs, governance technologies, BI estate and cloud investments remain important, and most of them are doing exactly what they were bought to do. Our advice to customers is consistent. Start with the decision, work backward, and do not rip and replace what you already have. Activate it.

The part I would draw attention to is this. With Latttice, governance travels with the data product. That is more than governance being applied once at the moment of creation. The appropriate policies and controls are embedded with the product and continue to operate as that product is appropriately reused and consumed.

In the environment Gartner describes, that distinction becomes increasingly important. A person can consume the product. Analytics can consume it. An application can consume it. AI can consume it. An agent can consume it. The consumer changes, and the organization does not have to reconstruct the governance from scratch each time.

The product travels. The context travels. The governance travels with it.

Active Governance Is What “Operationalize” Means

This is where Active Governance becomes more than a governance philosophy. If governance travels with the product, the controls can operate in the interaction itself. That matters increasingly as the consumer shifts from a person to a machine.

An experienced analyst will sometimes catch a bad result because they know the business, know the history of a metric, or know that one region reports differently. AI cannot be expected to rely on organizational folklore. Whatever context and control the interaction requires has to arrive with the information, which is the practical meaning of Gartner's phrase about operationalizing governance.

Trusted Data for People and AI

Gartner's emphasis on trusted, context-rich data being accessible to both humans and AI agents lands on an architectural problem we have been working on for some time. Organizations should not need one version of trusted enterprise understanding for people and another, disconnected version for AI. That path leads to two estates that drift apart, and to exactly the inconsistency that erodes confidence in both.

A trusted data product should be capable of becoming a reusable source of organizational understanding for appropriate consumers. A business person may use it to make a decision. Analytics may consume it. An application may consume it. AI may consume it. An agent may consume it. The important distinction is that the trusted business understanding does not have to be reconstructed for every new consumer.

That is where reusable data products become particularly important for enterprise AI. Not because reuse is efficient, although it is, but because reuse is what keeps a single, governed version of business meaning in place as the set of consumers expands.

Where Lenz Extends the Possibility

If Latttice creates the trusted, governed, context-rich data product, then Lenz allows organizations to build AI agents that operate on those trusted products. It is the logical next step rather than a separate capability, because an agent is only as trustworthy as the information and governance underneath it.

Building an agent is increasingly easy. Giving an agent trusted organizational understanding is much harder, and that is where Latttice and Lenz become strategically interesting together.

This progression changes the question leaders should be asking about enterprise AI. The question is no longer simply whether we can build an agent. Most organizations can now build an agent. The more important questions are what the agent knows, what trusted data it is operating on, what that data means to the business, what it is permitted to access, what governance applies, and what decision it is helping someone make or being permitted to make itself.

Data Products as Reusable Organizational Understanding

It would be a mistake to describe reuse purely as a technical efficiency. Reuse is strategically important, because every trusted data product captures a piece of organizational understanding: what information matters, what it means, how it relates to other information, what business logic applies, what governance applies, and the purpose for which it can appropriately be used.

When that product is reusable, the organization does not have to rediscover that understanding every time another person, application or AI system needs it. Over time, trusted data products begin to create a reusable layer of organizational understanding across the enterprise. That is a very different asset from a well-populated catalog.

This is where Gartner's emphasis on semantics and context matters most. AI does not simply need more enterprise data. It needs more enterprise understanding. That is a much bigger distinction, and it is the one I would encourage leaders to hold onto as they evaluate their own AI programs.

The Decision Remains the Anchor

None of this technology discussion changes the principle we keep returning to. The end point is the decision. We recommend that customers begin with the decision and work backward, and that principle applies equally to people and to AI.

Do not start by asking where we can deploy AI agents. Start by asking what decision we are trying to improve. Then work through what information that decision requires, what business context is required to interpret it, what governance applies, what trusted data product would support it, and who or what should be permitted to consume that product.

Only then does the automation question become useful. Should the decision remain human-led? Should AI assist? Should an agent recommend an action? Should an agent eventually be permitted to execute it? Answered in that order, those questions are governable. Answered in reverse, they become a series of pilots looking for a purpose. This is the progression from data-driven to Decision-Driven.

The Bottom Line