Judgement is the scarce resource

When something becomes cheap, something next to it becomes valuable. When calculation became cheap, asking the right question became scarce. AI is making answers cheap: the drafting, summarising and predicting that fills office work. What becomes scarce is judgement.

The age of cheap answers

Is this lead worth a call? Which project will slip? Answering such questions used to take a person's time. Now a system can read every email in a pipeline and flag the deals at risk, at a cost that keeps falling. When answers are plentiful, the bottleneck moves: to deciding which answers to act on, and standing behind the choice.

What happened to the bank teller

As cash machines spread in the 1970s and 1980s, the obvious prediction was that bank tellers would disappear. The economist James Bessen, among others, has shown that for a long time they did not. Cash machines made branches cheaper to run, so banks opened more of them, and the teller's job shifted from counting notes to explaining products and building relationships.

Automation sometimes does remove jobs, and pretending otherwise helps no one. But when a machine takes the routine part of a role, the value of the rest tends to rise, and the organisations that gain invest in that remaining part.

What judgement is, and is not

Intelligence is reaching a sound conclusion from information, and machines are getting good at it. Judgement is deciding what to do when information is incomplete, the stakes are real and someone must answer for the outcome.

That last element is the one a machine cannot supply. Accountability needs someone who can be asked why, who bears the consequence and who learns from it. That is why agents do the work and people keep the decisions: not a claim about how clever machines are, but about whom a customer or a court can hold responsible. Old wisdom, new intelligence: neither is enough alone.

A model can recommend a discount. It cannot be accountable for having given it.

The case against

Machines already make consequential decisions: credit scores, fraud flags, airline seat prices. Why not discounts? Because those systems work within rules people designed and remain accountable for. The machine decides the case; people decide the rule. And when automated decisions go badly wrong, it is often because no one noticed the rule no longer fitted the world, which is itself an act of judgement.

Does human approval slow everything down? It can, if the line is misplaced. Put it only where the business is committed, and let everything below run freely.

Growing more judgement

Much judgement was learned by doing routine work that machines now take on. The young salesperson learned which deals were real by updating a hundred of them. That apprenticeship now has to be replaced on purpose: people learn by explaining why a machine draft is wrong, managers make their reasoning visible, and decision rights widen as people show good judgement.

Mid-sized firms have an advantage here. The distance from a junior employee to the chief executive is short, so reasoning can be shared in the room. Reviewing a hundred machine drafts with a thoughtful manager may teach faster than writing ten by hand, but only if someone designs it.

What this means for a mid-market leader

Map your decisions, not only your processes. List the decisions that commit your business and name who makes each one.

Invest in the people who decide. Put some of what automation saves into training, context and time to think.

Protect the apprenticeship. Junior people must still learn why, even without writing the first draft.

Measure decisions, not just output. When options multiply, the quality of choices separates good companies from ordinary ones.

Answers get cheaper every year. The person who will stand behind a decision does not. Build around that, and machines will make your people more valuable, not less.