For years, we’ve heard the mantra “shift data governance left.” This means starting governance early in the data lifecycle and embedding quality, compliance, and understanding as close to data creation as possible. While well intentioned, this mindset oversimplifies how data actually works in the wild.
The reality is that governance isn’t a linear process—it’s a loop. Treating it like a one-way operation overlooks the most critical element: the real-time, context-rich questions users ask when they interact with data.
The pitfalls of linear governance
Most data governance strategies are designed around a production pipeline, under the assumption that data is created, documented, validated, and then consumed. In reality, businesses often consume largely undocumented data after it’s created, without pausing to validate it first. The linear governance approach overlooks the complexity of work that happens at the consumption stage. Data analysis and insights, redefinitions, and misunderstandings all happen at the point of use—and that’s where governance should begin, not end.
In practice, we see that data governance is driven by how it’s used. Someone asks, “What does this field mean?” or “Why is this number different from last month?” Those questions prompt inquiry. They uncover missing metadata, inconsistent definitions, or gaps in lineage. They spark change upstream.
In that sense, good governance doesn’t just shift left. It loops. It improves over time with inputs from data usage. If you’re using a data catalog but there’s no accommodation for this feedback loop, your data catalog will quickly become a stale repository of outdated, inaccurate data definitions.
Catalogs should answer questions, not create them
Another common misconception is that users need to be trained to use data catalogs. If your catalog requires extensive training in order for anyone to get value out of it, chances are it won’t be widely adopted, if at all.
Data discovery should feel natural and intuitive. It should begin with a question, not with a training manual. This is the premise behind Coalesce Catalog. We’ve designed it to be answer-first, allowing non-technical data users to interact naturally without needing to navigate deep hierarchies or learn new tools. Business users can ask questions like “What was our churn rate in Q4?” or “Where does this metric come from?” and their catalog should instantly direct them to the most appropriate, accurate dashboard or help them arrive at the right SQL to get that answer.
Everyone should be able to get a clear, trusted answer easily and not have to search through a list of numerous tables with similar names to find the one they need. Or be able to ask for a definition or lineage, and get a response from their data catalog like they would from a data engineer.
The same goes for technical data users. They don’t want to spend time searching through countless schemas or writing detective-level SQL just to find the right table. Coalesce Catalog streamlines that experience by surfacing definitions, lineage, and usage patterns with one query. It puts the intelligence where the user is, not buried behind layers of UI. Navigating a metadata tree or running exploratory queries just to find the right table will slow anyone down.
This is where agentic AI comes into play—and it doesn’t just enter the chat, it rewrites the script.
The role of agentic AI in non-traditional data governance
Think of the role of agentic AI in data governance and discovery like having your own personal chef. You don’t need to know what’s in the pantry or how to prep ingredients. You say, “I want spaghetti,” and the chef handles everything, from shopping to cooking to serving.
Agentic AI does the same for data—it doesn’t just search metadata, it understands intent. Agentic AI can:
- Infer the best table or metric based on usage patterns
- Resolve ambiguity in natural language queries
- Automatically pull in relevant documentation, lineage, and ownership context
- Learn from previous interactions to improve over time
Under the hood, this requires real orchestration. It involves embedding large language models (LLMs) with vector search over metadata, integrating usage logs to prioritize commonly queried entities, and applying fine-grained access controls on the fly. It’s not just a chatbot bolted onto your data. Instead, it’s an intelligent system that responds like a human, but scales like software.
Closing the loop with Coalesce
When AI acts as an agent within the context of your data catalog, it closes the governance loop. As users interact with the system, their questions become part of the knowledge graph—enhancing documentation, improving discoverability, and training the AI to serve better answers over time. At Coalesce, we’re making sure that our catalog doesn’t just provide data, but that it also learns from usage to build a smarter, more adaptive governance ecosystem. Questions downstream generate context that improves documentation, reveals issues, and trains the model. It becomes a virtuous cycle: the more people ask, the better the system gets.
Let’s move beyond the outdated idea of shifting governance left. Instead, we should build systems that continuously listen at the point of use, capture insight where it’s generated, and intelligently propagate that knowledge back through the data lifecycle—closing the loop in real time.
See Catalog in action
Check out our latest Demos with Doug session with Tristan Mayer, General Manager of Coalesce Catalog. In this video demo, Tristan covers AI-driven automation, built-in governance, repeatable transformation patterns, and performance optimization—all designed to accelerate your AI and analytics initiatives with ease.
Interested in seeing what AI-powered data cataloging can do for your team? Book a demo with us to learn more.