Why AI did not fix your data governance

Somewhere in the last two years, you or a steward on your team opened a general AI chat and asked it to help write a data quality rule. The answer was good. Fast, articulate, better than the blank page. And yet, a year later, your governance is roughly where it was.

Somewhere along the way, the story also flipped from the tool will fix it to AI will fix it. Same trap, new host. AI on its own fixes nothing. It is leverage, and leverage needs something to push against: the underlying knowledge, the guidance, and the architecture that put the right tool on the right job at the right time.

This is not an article telling you AI was the wrong idea. I build with these models daily, and the honest version is more useful than the hype in either direction: general AI is a genuinely strong conversation partner and a structurally poor governance system. Knowing exactly where the line sits is worth real money.

What it is genuinely good at

Put one element in front of a general model and ask it to think with you, and it delivers. It drafts a plausible data quality rule. It challenges a definition. It lists edge cases you had not considered. For a steward with nobody to bounce reasoning against, that alone beats working alone.

If that were the whole job, the problem would be solved. It is not the whole job. It is the easiest third of it.

The three gaps that do not close with a better prompt

It does not hold a method. Good governance work has an order: what breaks, who feels it, what kind of risk this is, and only then what to build. A general model follows whatever order your prompt implies today. Tired steward, thin prompt, thin reasoning. The method lives in your head and travels with your energy level, which is exactly the problem you had before AI.

It gives different answers to different people. Two stewards, same element, same company, different prompts, different rules. Ask a governance team what they need and consistency is near the top of the list. A tool that reasons differently for every user on the same problem is generating the numbers do not match meeting, one artifact at a time.

The work evaporates when the chat ends. An hour of good reasoning surfaces five follow ups: an owner to confirm, a definition to settle, a condition to verify. Close the tab and every one of them now lives in your head. Nothing is tracked, nothing is connected to last month's session, nothing reminds you. The thinking was real. The body of work never existed.

The tool helped you think for an hour. The program did not move.

The test to run on any AI in your governance stack

Three questions, in order. Does it hold the method, or do I have to bring it in every prompt? Do two people on the same problem get the same reasoning? And when the session ends, does the work persist somewhere that is not my memory?

The harder part

The uncomfortable truth is that these gaps are not flaws in the models. They are the definition of general purpose. A tool built to be anything for anyone cannot also be a held method, a consistent reasoning path, and a persistent body of work for one specific discipline. That takes the method being built into the structure, not pasted into the prompt.

That is the difference enabledat is built on: the consequence first method held as a structured reasoning process, the same path for every steward, and every session feeding a tracked body of work instead of evaporating at the end of a chat. The model does the thinking with you. The structure makes it governance.

Use general AI for what it is, a sharp conversation. Just stop expecting a conversation to become a program on its own.