AI in Data Management: A Reality Check on the „Agent Wonderland“ [EN]

Everywhere you look, AI agents and features are being touted as the next frontier in data management. The promises are grand: complete automation, hyper-intuitive user experiences, and self-optimizing systems built in hours, not days. This is the „agent wonderland“ we’ve been sold. But as a Senior Analyst at BARC, I’ve learned to look past the hype.

To understand the true state of affairs, I challenged my perceptions by doing a small study, interviewing the people on the front lines: the software vendors building the tools, the service providers implementing them, and the end-users trying to derive value from them. My findings paint a clear picture:

While the AI revolution in data management is coming, we are not in a wonderland just yet. In many ways, we’re still in an „AI kindergarten,“ learning the rules of the playground.

No 1: Complex AI Transformation is Rare; We’re Still Mastering Simple Tasks.

The research shows that despite the buzz, truly transformative, complex AI use cases are the exception, not the rule.

The majority of AI adoption isn’t about overhauling entire business processes. Instead, it’s focused on highly individual implementations designed to automate singular, well-defined tasks.

The most common and mature applications of AI in data management today are functional and targeted, primarily benefiting technical users like data engineers.

These include:

• Coding Assistance: Features like code documentation, explanation, reverse engineering, and co-pilot support are the most widely used.
• Text-to-Code: Translating natural language into SQL and other code is a popular and practical application.
• Data Quality: AI is being used for tasks like generating data quality rules, detecting anomalies, and identifying duplicates.
• Metadata & Discovery: AI helps with semantic discovery, generating metadata descriptions, and classifying PII.

While these are valuable efficiency gains, they are a far cry from the vision of autonomous, self-governing data ecosystems. The current landscape is dominated by these specific, isolated functions rather than integrated, strategic AI deployments. This points to a market that is still experimenting and learning to walk before it can run.

No 2: The Great Divide – The Management Attention vs. Execution Gap.

There is an undeniable demand for AI. My impression and also the study found that management, in particular, is driving the call to „do something with AI.“

I asked my 11 interview partners for their assessment of how high the demand for AI is in companies. On a scale of 1 (low) to 10 (high), the demand for AI in data management was rated a strong 6.8, with vendors seeing it as high as 8.0. However, this enthusiasm is running headfirst into a harsh reality: a profound lack of readiness. I identified a critical „management attention-execution gap.“

While executives is is delighted with simple functions, expect immediate transformation, the foundational work required to make AI effective is lagging far behind. The interview partner confirmed and rated the average company’s maturity to apply AI in data management at a sobering 3.6 out of 10.

This gap is built on several key challenges:

• Foundational Infrastructure Gaps: Most organizations lack the mature AI/data governance, metadata infrastructure, and what I call the „context and knowledge layer“ that are prerequisites for effective AI.

Without business context, process knowledge, and reliable data, even the best agent can only produce generic, or worse, hallucinated, output.

• „Quick and Dirty“ Dominance: In the rush to innovate, companies are favoring coding scripts and open-source tools over systematic, governed data management. This creates a „Wild West“ of AI applications, stamping out siloed solutions that are fast locally but lead to enterprise-wide chaos and loss of control. That’s a big risk.
• Strategic & Cultural Deficits: Beyond technology, there is a lack of strategic orientation, AI skills, and a Data & AI Culture to support a sustainable AI strategy.

The bottom line is that you cannot build an „agent wonderland“ on a weak foundation. The ability to build, manage, and protect this context and knowledge layer will become the key competitive differentiator of the future. It is your company’s intellectual property, and it must be treated and enhanced as a strategic asset.

No 3: The Tools Are Not Ready – Welcome to the „Beta World.“

Over the past few months, I’ve had the opportunity to test out various software programs. Some of the AI features are really cool and give you a glimpse of the potential that lies ahead. It’s kind of like discovering gold dust and now being on the verge of finding the mine. Enthusiasm everywhere.

Given the gaps in company readiness, one might hope that standard data management software could provide the necessary guardrails. But the research (and also the result of my software tests) shows that the AI features within these tools are still in their infancy, or just rolled out to selected customers.

Service providers delivered a particularly damning verdict, stating that the features are „not at a level where you can build professional environments.“

The overall maturity of AI features in existing tools was rated a mere 3.4 out of 10. The primary complaints centered on a lack of flexibility, functionality that is inferior to specialized models, and simple feature availability issues.

Figure 3: Service providers are not satisfied today, and vendors see considerable potential for improvement. Given the pace of innovation, this figure may already be outdated. Number to the left: service provider, number to the right: vendor. Number above is a total.

This immaturity has created a dangerous vacuum. Because the standard tools are insufficient, companies and service providers are forced to build their own solutions.

This fuels the „quick and dirty“ approach, where governance and control are often sacrificed for speed and flexibility. We’ve seen this movie before with the uncontrolled proliferation of Excel and self-service analytics. However, the risk is now exponentially greater when anyone with a large language model can create an „agent,“ leading to spiraling costs, diminished output, and hallucinating bots.

We must bridge this gap between open-source flexibility and enterprise governance. Standard data management tools are essential for bringing these two worlds together, especially when it comes to unifying context, knowledge, and metadata into a single, reliable source of truth for both humans and agents.

And I have to put that immaturity into perspective and say that for simple use cases, the AI functions already deliver added value in terms of productivity and applicability, e.g. identify, generate, test dq rules, create first decriptions for data products or assets, identify sensitive data or support data classification, analyze unstructured data and excerpt metadata and more.

The Path Forward: From AI Kindergarten to Governed Experimentation

The conclusion is clear: we are in a dynamic, exciting, but messy experimental phase. To move forward, we must graduate from the „uncontrolled AI kindergarten“ to a world of governed experimentation in safe sandbox environments. And from there to a fully governed production environment.

The path to a mature, AI-driven data management landscape requires a focus on several key imperatives:

1. Embrace Governance-by-Design: Governance cannot be an afterthought. It must be woven into the fabric of your AI strategy from day one.
2. Understand the Importance of Linked Knowledge and Context. AI doesn’t operate in a vacuum. It requires a deep understanding of your business—your processes, your rules, your data, and the relationships between them. This linked knowledge and context layer is the fundamental prerequisite for any successful AI implementation. Without it, you are simply automating tasks without intelligence. Building and actively managing this foundation is the most critical first step.
3. Elevate the Knowledge Worker: The focus must shift from simply engineering data pipelines to engineering knowledge and context. The „Data Engineer“ must evolve into a „Knowledge Engineer.“
4. Keep Humans in the Loop: Trust is the foundation of AI adoption. Human supervision and intervention remain critical for validation, risk control, and building confidence.
5. Protect Your Context: Your organizational context is a strategic asset. It is your unique intellectual property and must be curated, enhanced and protected. This concept of „Context Sovereignty“ is paramount.

The „agent wonderland“ is not a pre-built theme park we can simply buy a ticket to. It’s more than that. The use of AI demands transformation. We can’t just throw AI at our existing processes and automate the junk. Instead, a targeted AI strategy requires us to fundamentally rethink our processes in light of today’s AI capabilities, and to align the foundational elements. It’s something we must build ourselves, piece by piece, on a solid foundation of governance, context, and strategic foresight.

Methodology

The methodology is quite simple. I’ve generated a few leading thesis about AI in Data Management out of my experiences and talks to data & analytics end-user companies. I created a short questionnaire that was used as a structure for a 60-min interview. I did 11 interviews in total. One with each of the service provider / vendor listed below. I summarized each interview and compiled the results. Finally, I cross-checked the results with BARC research studies like the Data Preperation for AI report published in March 2025.

This blog is a summary of the key insights.

Many THANKS to all my interview partners who took the time to discuss this exciting topic with me.