Looking back, 2025 was not the year enterprises “figured out” AI. It was the year many organizations finally admitted how unprepared they were.

When generative AI burst onto the scene in 2023, adoption was fast and furious. Off-the-shelf tools promised immediate gains in productivity for writing, coding, and summarization; organizations embraced them enthusiastically. At the time, there was a lot of discussion that companies would begin to use their own data with generative AI, starting with text and eventually expanding to structured data. Text-based use cases did take hold in late 2024 and into 2025, but broad use of enterprise data remained limited.

The reason became increasingly clear in 2025: organizations still needed more time to get ready for AI. That may be because a large percentage of them never had their data foundations in place for either analytics or traditional AI.

Throughout 2025, data management for AI emerged as the top priority by a wide margin. Companies recognized that AI can stall quickly without trusted, accessible, and well-governed data. At the same time, the explosion of consumerized generative AI tools led directly to concern about guardrails. Shadow AI spread faster than policy, and many organizations in TDWI surveys reported taking AI governance more seriously as a direct response.

We also saw tangible progress. Generative BI gained traction as vendors and internal teams built natural language interfaces. And agentic AI moved from theory into early experimentation, with early adopters starting to experiment.  Still, the pattern was familiar: excitement outpacing readiness.

That brings us to 2026.

As I look ahead, I see 2026 as the start of a potential inflection point, particularly for agentic AI, but not the year it becomes mainstream. History supports this. Predictive analytics followed a similar path. For years, organizations said they planned to implement it. If intent translated directly into execution, adoption would exceed 80% today. Instead, I see in my research at TDWI that real implementation sits closer to half. The gap wasn’t technology, it was data readiness, skills, governance, and operational maturity.

Agentic AI raises the bar even higher. Agentic systems don’t just analyze or recommend; they act. That means organizations must trust the data, the models, the workflows, and the boundaries placed on autonomy. This is why agentic AI will unfold over multiple years, not one. In 2026, what we will see is the beginning of that journey.

The signals are real. In 2025, vendors heavily promoted agentic AI, and there is always a lag between vendor innovation and enterprise adoption. Anecdotally, I heard from numerous organizations that were already refining workflows in anticipation of agents performing discrete tasks.

At the same time, agent orchestration is emerging as a differentiator. Coordinating multiple specialized agents to accomplish complex tasks is no longer hypothetical, and standards such as Model-Context-Platform (MCP) are gaining attention as foundational components. Still, most organizations will move cautiously. Just as with machine learning, early adoption does not translate into immediate scale.

Data Foundations and Governance are key

What will determine whether this inflection point leads to real progress, or stalls, is data and governance.

Many organizations are already discovering that consumerized generative AI has a value ceiling. While the vast majority of organizations we survey at TDWI claim to use generative AI, fewer than half have implemented machine learning. Copilot-style tools deliver convenience, not sustained advantage. Our research shows organizations relying primarily on these tools are less likely to generate significant value than those using enterprise data.

Chatbots illustrate the problem. A chatbot that cannot access customer history, operational systems, or structured data quickly becomes superficial. Maintenance bots that rely only on manuals and FAQs are useful, but limited. Real impact requires company data, integrated systems, and strong foundations.

This is where AI governance becomes unavoidable. Governance for AI, distinct from traditional data governance, focuses on trust, transparency, accountability, and oversight. With agentic AI, the questions multiply: What should agents have access to? What happens during handoffs? How do you audit decisions made by a group of agents? Without lineage, access controls, and assurance frameworks, organizations will not move beyond experimentation.

So, What About 2026?

Finally, 2026 is when unstructured and multimodal data truly become first-class citizens. Text data is already mainstream, and images and video are following quickly. Organizations are investing in lakehouses that support multiple data types, and early generative AI use cases have demonstrated the value locked in unstructured data.

Taken together, 2026 is perhaps best understood as a year of building—not completion. Organizations will begin to use enterprise and unstructured data more seriously for AI. They will start implementing AI governance with intent. And they will cautiously advance toward agentic AI, recognizing that autonomy without data and governance is risk, not innovation.

The inflection point is real—but only for organizations willing to do the foundational work required to cross it.