How do you scale a person’s judgment?

There are two ways to make a busy person appear more productive. One is to fit more into the calendar. The other is to make some of the calendar unnecessary.

The second interests me more. It requires understanding what keeps returning to the person: a judgment nobody else feels able to make, a relationship nobody else can carry, or a commitment that never quite closes without intervention.

“Scale the leader” sounds grand until it becomes specific. Which recurring decision could someone else make well? What would they need to understand? What authority would they need? What must still come back?

The correction is part of the evidence¶

Building this notebook gave me a small case to inspect. A first version produced functioning pages, but I still had to clarify public access and ask for more of my actual work and original figures. I am recording that gap in the personal-AI experiment log. It is an early observation about delegation, not a demonstrated productivity gain. The assistant needed a more precise account of the outcome, not simply more information about me.

More information can create more work¶

An executive can receive a superb daily briefing and still have the same problem. Now there is another document to read before deciding what to do.

Simon treated attention as a scarce organizational resource, and argued that an information-processing system helps only when it reduces the demands placed on the rest of the organization.¹ That suggests a better standard for support work: how much relevant complexity has been absorbed, and how much attention is still required to reach a sound decision?

A short memo can fail this standard if it hides the decisive uncertainty. A longer one can pass if it settles a complex question that would otherwise require five meetings. Brevity is useful; the amount of unresolved work transferred to the reader is the more important variable.

My preferred output is a recommendation with its reasoning, strongest objection, and requested decision. If no decision is needed, say what was done and what remains open. Do not make the reader reverse-engineer the purpose of the update.

Learn the reasons, not just the answer¶

Suppose a leader rejects an opportunity. A weak record says “pass.” A better one captures the concern: unreliable evidence, an unattractive distribution channel, a team that does not learn, or a price that assumes too much.

Those are different lessons. Recording only the verdict invites imitation of superficial features. The next opportunity may look similar while differing on the variable that mattered.

I would begin an apprenticeship with independent assessments. Before hearing the principal’s answer, the apprentice writes a recommendation and the assumptions behind it. Then the two compare reasons, not merely agreement. Otherwise hindsight makes every decision seem obvious and conceals where the apprentice’s model diverges.

The record should include uncertainty. “I do not know enough about this domain” is different from a confident negative. It should also include the alternatives that were rejected, because future learning often arrives as evidence that the discarded path was better.

Give authority a shape¶

Understanding someone does not automatically authorize acting for them. A useful delegation specifies the decision class, resource limits, commitments that cannot be made, and conditions for escalation.

Bezos’s discussion of reversible decisions is a helpful starting distinction.² It is a management heuristic, not a reason to apply the same speed standard to every domain. A reversible scheduling change and a consequential external promise do not belong in the same bucket.

I would expand authority through a sequence: prepare a recommendation; carry out an agreed decision; own a bounded recurring decision; then represent the principal in situations where both the mandate and the exceptions are clear. The boundaries can grow as evidence of judgment accumulates.

Importantly, the person receiving a delegated conversation should know the scope of the delegation. Ambiguity can make an assistant look powerful while leaving everyone unsure what has actually been committed.

Do not automate someone’s blind spots¶

The difficult part is that an exceptional person is not exceptional at everything. Kahneman and Klein argue that the quality of intuition depends on the predictability of the environment and opportunities to learn its regularities.³ Confidence is not an adequate substitute for those conditions.

An apprenticeship should therefore preserve the ability to disagree. A documented preference is evidence about the principal. It is not proof about the world. When a familiar heuristic enters an unfamiliar domain, the system should become more inquisitive, not more obedient.

The most useful disagreement brings a better account of the problem. It names the assumption, presents contrary evidence, and offers an alternative with its own costs. Mere contrarianism consumes attention just as surely as uncritical agreement.

Measure independence with correction¶

Fewer escalations would be an incomplete success metric. People can reduce escalation by concealing problems. Faster decisions can produce more rework. More meetings handled independently can mean more relationships quietly damaged.

I would pair measures of autonomy with measures of correction: decisions completed without unnecessary intervention, subsequent reversals, avoidable rework, unresolved commitments, and how quickly material surprises reached the principal. Review a sample of ordinary decisions as well as the obvious failures; otherwise the evaluation sees only what gets reported.

This is a proposed operating system, not a proven universal protocol. I would test it on one recurring class of work before expanding it. If documentation takes more time than it saves, simplify it. If decisions remain dependent on constant clarification, examine what the records are missing.

The ambition is an organization that can extend good judgment and correct bad judgment. A leader has truly gained capacity when useful work reaches a sound conclusion without requiring their presence at every step.