Judgement has to become infrastructure in the AI era, plainly, because the informal version, good judgement quietly held in the heads of experienced people, was never designed to operate at the speed AI now allows decisions to move. Infrastructure means judgement is deliberately built, explicitly owned, and visible enough that someone outside the room could ask how a decision was made and get a specific answer, not an assurance that the right people were probably involved.
Why Informal Judgement Worked Until It Didn't
This point is genuinely worth taking seriously rather than dismissing outright as an abstract concern. For most of a business's history, judgement being informal genuinely wasn't a problem, because the pace of decisions was naturally bounded by how fast a human could think through them. That natural bound acted as a built-in safety margin. A leader with good instincts could carry an enormous amount of unwritten judgement in their head and the business would run fine, because the slowness of human reasoning gave everyone time to notice if something was drifting off course. AI removes that bound. Decisions that used to take a day now take minutes, and the safety margin that came from sheer slowness disappears along with the slowness itself, which means whatever judgement used to live safely in someone's head now has to live somewhere faster, more visible, and less dependent on that one person happening to be paying close attention in the moment.
The Core Shift: Judgement used to live quietly in people's heads, applied case by case, at human speed. At AI speed, it has to live somewhere everyone can actually see it, named, written down, and checkable after the fact.
Why Judgement Has to Become Infrastructure, Not Just a Good Habit
The distinction here genuinely matters far more than it first sounds on the surface. A good habit is something an individual does consistently because they're conscientious, and it disappears the moment that individual is unavailable, distracted, or replaced. Infrastructure is something the business depends on regardless of which individual happens to be present. Turning judgement into infrastructure means three things exist in writing, not just in someone's reliable habit: a named owner accountable for each category of AI-assisted decision, real and exercised authority for that owner to overrule an output, and a standard specific enough that a decision could be checked against it after the fact, by someone who wasn't in the room when it was made.
- Named, Not Assumed: A specific accountable person exists for each decision category, not a general sense that someone competent is probably watching.
- Exercised, Not Theoretical: Override authority is actually used often enough to prove it's real, not just technically possible in principle.
- Checkable, Not Vague: The standard a decision has to meet is specific enough that someone outside the original decision could audit it afterward.
I've genuinely watched leadership teams resist this exact framing at first, almost every single time, because it can sound like an accusation that their people can't be trusted. It isn't. The people are usually fine. What's missing isn't trustworthy individuals, it's a structure that doesn't depend entirely on those individuals staying attentive, staying in the seat, and staying available at exactly the moment a decision needs their judgement. Infrastructure isn't a statement of distrust in your people. It's an acknowledgement that trust alone was never actually a system, it was a hope that the right person would be paying attention when it mattered, and hope is not something a business should be depending on for its highest-stakes decisions.
Why Boards Are Now Asking for This Directly
The shift is genuinely recent, and it's worth being quite precise about exactly what caused it rather than treating it as a vague, purely ambient trend. As AI-assisted decisions have moved from experimental pilots into genuinely material parts of how businesses operate, the potential downside of an unmanaged failure has grown alongside them, and boards, whose entire function is managing exactly that kind of downside risk, have started asking sharper, more specific questions as a direct result. This has genuinely stopped being a purely internal operational question, and it's worth understanding exactly why the shift has happened now rather than earlier. Board members increasingly ask leadership teams, directly, how AI risk is being managed, and the honest answer from most organisations is still a version of general reassurance, we have good people, we're being thoughtful, we're monitoring the situation closely. That answer used to be sufficient. It increasingly isn't, because a board member who's seen a peer company face a genuine AI-related failure knows exactly what question to ask next: who specifically owns this, and what happens when they disagree with what the system produced. A leadership team that can answer with a name, a specific authority, and a specific standard is in a fundamentally stronger position than one that can only offer general confidence.
I'd genuinely go further and say this specific distinction is fast becoming one of the clearest external signals available of whether a leadership team has done real work here or adopted the vocabulary of responsible AI use without the structure underneath it. Vague reassurance is cheap to produce and increasingly transparent to a board, an investor, or a regulator who's learned what to listen for. Named ownership, exercised authority, and a checkable standard are much harder to fake convincingly, which is exactly why they're the right things to build regardless of whether anyone outside the business is currently asking about them yet.
- Write down what currently only lives in someone's head — Start with your highest-stakes AI-assisted decision category and make its judgement explicit: who owns it, what authority they have, what standard applies.
- Test whether override authority is actually used — If nobody can recall a specific recent instance of overruling an AI output, the authority is theoretical, not infrastructure, regardless of what the policy says.
- Make the standard specific enough to audit — "Use good judgement" isn't a standard anyone could check a decision against afterward. Write something concrete enough that a different person could evaluate the same decision.
- Prepare the board-level answer before you're asked — Draft the specific, named answer to "who owns AI risk in this decision category" now, rather than improvising a version of it under pressure in a board meeting.
There's a genuinely practical sequencing question worth answering honestly too, before any real redesign work actually begins in earnest: which decision category should get judgement infrastructure first. The instinct is often to start with whatever feels most visible or most urgent this quarter, but the better starting point is whichever category carries the combination of highest stakes and highest current AI involvement, because that's where the gap between informal judgement and the pace of decisions has already grown the widest. Starting there, even if it's uncomfortable, tends to produce the clearest proof of concept, and the clearest argument for extending the same infrastructure to the rest of the business afterward.
The Cost of Leaving Judgement Informal
The real cost here isn't usually one single dramatic, catastrophic event, which is exactly what makes the informal version so easy to leave unaddressed for so long. It's a slow accumulation of small exposures: a decision category where accountability has quietly thinned as volume increased, a leader who intended to stay closely involved but whose actual review became a glance rather than genuine scrutiny as other priorities competed for their attention, an override authority that technically exists but that nobody has actually exercised in months because doing so has quietly started to feel like second-guessing a system that's usually right. None of those individually looks alarming. Together, over time, they're exactly the conditions a genuine failure needs to happen with nobody clearly able to explain afterward how it was allowed to.
There's a genuine cultural dimension to all of this too, and it's one worth naming directly and honestly, and it's worth naming because it's often the real obstacle rather than the mechanics of building the infrastructure itself. Some cultures genuinely prize speed and autonomy, trusting people to move fast without asking permission, and that culture can experience judgement infrastructure as a step backward, more process, more sign-off, more friction on exactly the thing the culture is proud of. The businesses that navigate this well don't abandon the culture. They build infrastructure narrow enough to cover the genuinely high-stakes decisions without touching the vast majority of everyday calls the culture's speed was always meant to apply to, which keeps the autonomy where it belongs while closing the gap where it actually matters.
The Distinction That Actually Matters
Judgement as infrastructure isn't about distrusting your people or slowing down decisions that don't need slowing down. It's about recognising that the informal version of judgement, however excellent the people holding it, was built for a pace of decision-making that AI has already left behind. Build the infrastructure deliberately, name the owners, exercise the authority, write the standards down, and judgement survives the speed increase intact. Leave it informal, and speed will eventually outrun whatever quiet, unwritten judgement was holding the business together.
The question worth sitting with is a simple one: if a board member, an investor, or a regulator asked tomorrow who owns judgement over your highest-stakes AI-assisted decision, could you name them, describe their authority, and point to the standard they're holding the decision to. If the honest answer is no, that's not a failure. It's simply the starting point, and it's a far better place to start from deliberately than to discover the same gap for the first time in the middle of an actual crisis.