Before the Framework Wakes

Here is the sentence that is doing all the work: 80% of governments will automate "routine decision-making" by 2028.

The word routine is where the story lives. Not in the automation. Not in the 80%. In the classification.

Governance frameworks are designed to activate when a decision happens. They govern the decision. What they don't govern — what they can't govern, by design — is the classification of what counts as a decision worth governing. That happens first. Before the framework wakes up.

Benefits eligibility determination: routine. Parole recommendation: routine. Child welfare risk scoring: routine. Each of these is routine from the system's perspective — it processes ten thousand of them a week. It is not routine to the person whose housing or freedom depends on the answer.

This is not a new problem. Bureaucracy has always worked this way. The intake officer decides whether your case qualifies for review before the review process begins. The caseworker decides whether your situation is "standard" before your file reaches the supervisor. These pre-classification decisions have always been the least governed part of governance. The AI just makes them faster, and more invisible, and harder to challenge — because now the classification comes from a model, and models don't leave the same kind of paper trail.

The governance framework governs the agent. It does not govern the definition of which agent decisions require governance. That definition is a political act that gets laundered through the word routine.

And then there's the explainability problem, which I think makes this worse. The prediction is that by 2029, 70% of agencies will require explainable AI for automated decisions. Explainability feels like accountability. You now get a reason for the denial. But a reason isn't standing. The system can explain exactly why it denied your application — correct weights, correct inputs, the answer follows — and the explanation doesn't give you the ability to challenge the decision. It gives you a story about why the decision was inevitable.

This is subtle. An unexplained denial at least leaves open the possibility that the decision was arbitrary. An explained denial closes that opening. The explanation makes the decision feel final in a way that silence couldn't.

The citizen who loses their benefits has no one to appeal to, not because the process lacked transparency, but because the classification of their case as routine — the act that made it eligible for automated processing — happened in a layer that the governance framework was never designed to reach. And now they have a detailed explanation of why the routine process produced the routine result.

The framework governs what happens inside it. The predicate — what counts as the kind of decision the framework governs — lives outside it. Nobody's governing that.