Data Governance Engineering: Making Governance Invisible

Introduction: Governance That Moves at the Speed of Insight

"Good governance shouldn't slow you down; it should help you move faster with confidence." This is the work philosophy of Arnaud Gueulette, a Data Governance Engineer at Datashift. In the world of data governance, Arnaud's work proves that governance doesn't have to be a roadblock. Instead, it can be the invisible backbone that enables agility, compliance, and collaboration—without the friction.

For teams using platforms such as Snowflake, AWS, or Collibra, Arnaud's approach provides a blueprint for integrating governance seamlessly into operations. His methods focus on automation, Shift-Left principles, system thinking, and human-centric design, ensuring data remains reliable, compliant, and easy to use.

What Is Data (Governance) Engineering?

Data Governance Engineering is about balancing rigor with reality. It's not just about enforcing rules—it's about designing systems where governance feels like a natural part of the workflow, not an afterthought.

Core Principles:

1. Governance by Design: "I embed governance into pipelines from day one, so compliance becomes automatic." Arnaud uses Terraform and Terragrunt (IaC) to deploy platforms like Collibra DQ and Edge, ensuring governance rules are version-controlled, tagged and baked into infrastructure from the start.  
   

2. Practical Automation: "If it's repetitive, automate it." Using tools like AWS Event Bridge and Lambda, Arnaud builds event-driven workflows that handle metadata updates, trigger validations, and sync governance states across platforms—without manual intervention.
   

3. Shift-Left Governance: "Catching issues early saves time and headaches." By integrating governance checks directly into CI/CD pipelines, teams catch misconfigurations and policy violations before they break deployments—not after they reach production.
    

"The tools we have today are powerful, but the real magic happens when we use them to support, not slow down the team."

Model Context Protocol: Governance That Comes to You

The principles Arnaud follows—embedding governance early, automating the repetitive, making complexity invisible—all point toward the same goal: governance that doesn't interrupt flow. Recently, that vision found a new expression through the Model Context Protocol (MCP).

MCP is a standardized way for AI systems to talk to data platforms like Collibra and Snowflake. Instead of navigating complex interfaces or learning query languages, you ask questions in plain language and get clear answers.

"Think of it as a universal translator between people and their data tools," Arnaud explains. "It gives a ChatGPT-like assistant access to your governance context. Governance stops being a destination. It starts behaving like infrastructure."

How MCP Works: Two Complementary Modes

MCP operates in two ways, both designed to keep governance part of your workflow rather than separate from it.

Insight Mode is about discovery. "You ask questions in plain language, you get answers. No navigation, no new UI to learn." It helps you explore what you already have—data assets, metadata, lineage, ownership.

Agent Mode goes further. It lets you act through the same interface: update metadata, manage permissions, and trigger workflows. All through conversation, all governed by explicit policies that keep things safe.

"MCP doesn't replace what you're already using," Arnaud says. "It sits on top of your current setup and makes it easier to interact with governance without switching context."

A Real-World Example: Collibra MCP

At Datashift, Arnaud recently deployed Collibra's DQ, Edge, and EKS instances across dev and production environments, leveraging Terraform, Terragrunt, and CI/CD pipelines to ensure everything from infrastructure to policies is version-controlled, tagged and testable.

With that foundation in place, he tested the Collibra MCP—connected to an AI agent—across real governance scenarios.

"I could ask questions about data quality, metadata, lineage, business terms—all in plain language," he says. "No switching between screens, no hunting through the interface."

One question stood out: "What's the lineage for this customer dataset, and who owns the upstream sources?"

Normally, answering this requires navigating through Collibra's interface, opening multiple assets, checking relations, and cross-referencing business glossaries. Not hard—just friction-heavy. With MCP, the answer came back immediately: full lineage, data quality scores, related business terms.

"What surprised me wasn't just the speed—it was the completeness. The context I'd built over months—metadata standards, quality rules, business definitions—was suddenly accessible without having to know where to look for it."

But it wasn't just about discovery. He also generated outputs directly—data lineage reports, data quality risk assessments—without leaving the conversation.

"That's where it gets exciting. You're not just finding information. You're producing governance artifacts on the fly."

Challenges and Considerations

MCP is promising, but it's not without challenges. Arnaud sees a few areas that need careful attention:

• Security and Access Control. "When you give an AI agent access to your governance context, you need to be deliberate about what it can see and do." MCP inherits the permissions of your underlying platforms, but teams need clear policies on what actions Agent Mode can take—and who can trigger them.
  

• Data Quality as a Prerequisite. "MCP surfaces what's there. If your metadata is incomplete or inconsistent, that becomes visible fast." This isn't a flaw—it's a feature. But it means teams need to invest in data quality before expecting MCP to deliver value.
  

• Trust and Adoption. "People need to trust the answers they're getting." That trust comes from accuracy, consistency, and transparency about where the information comes from. Without it, adoption stalls.
  

• Maturity and Skepticism. "MCP is still new—and that means skepticism is natural." Not everyone is convinced yet, and that's fair. Technology needs advocates who can demonstrate real value, share early wins, and build confidence through practical examples rather than hype.

"These aren't reasons to avoid MCP—they're reasons to approach it thoughtfully. The foundation matters. We're already proving the concept with some of our clients."

Advice for Teams Starting with MCP: Build the Foundation First

"If I could give one piece of advice to teams just starting with MCP, it would be this: don't start with MCP, start with your foundations."

MCP isn't a magic fix for broken governance. It's an amplifier. If your metadata is inconsistent, or your access controls are poorly defined, MCP won't solve those problems—it will highlight them faster.

From a technical perspective, treat MCP as an interface layer, not a governance layer. Your existing tools—Collibra for metadata, AWS Lake Formation for permissions—should still enforce policies and controls. MCP makes that information easier to access through natural language.

"Think of MCP like a user-friendly dashboard for your governance engine. It doesn't replace the engine—it makes it easier to use."

Start with Insight Mode: let users ask questions like "Show me all datasets tagged as PII" or "What's the lineage of this report?" It's low-risk and high-value. Validate accuracy, refine the experience with a single team, then expand.

"Small steps lead to big improvements. The goal isn't to replace your governance tools—it's to make them work harder for you."

The Bigger Picture: Why Data Governance Engineering Matters

MCP is one tool among many. What matters more is the discipline behind it.

Data Governance Engineering sits at the intersection of infrastructure, policy, and user experience. It's about building systems that enforce the right rules without creating friction—systems that scale, adapt, and stay invisible until they're needed.

"The goal isn't to make governance the hero," Arnaud says. "It's to make governance disappear into the workflow. When it's done right, people don't even notice it's there—they just move faster."

Most people don't think about governance until something breaks. Arnaud wants to change that. "I want to build systems where governance is already there, working quietly, so teams can move fast without worrying."

As platforms evolve and AI becomes more embedded in data workflows, the demand for this kind of thinking is only growing. It's technical work with real organizational impact.

"We're not just configuring tools. We're shaping how teams interact with data. That's what makes this work interesting—and why I think more people should be doing it."

Conclusion: Governance That Works for You

Arnaud's work proves that governance doesn't have to be a bottleneck—it can be a catalyst. By integrating automation, human-centric design, and tools like MCP, teams can innovate with confidence, knowing their data is reliable, compliant, and easy to use.

"When governance is well-designed, teams can focus on what they do best—without worrying about compliance."

Ready to Transform Your Data Governance? Datashift’s team of Data Governance Engineers helps leaders build practical, scalable governance that works for their teams. Let’s talk about how we can support your data initiatives.