Lesson 0.4.10.4.1 Why judgment is the job

In everyday life. When phones began to remember every number, people stopped memorizing them. That was fine. Losing the ability to tell whether a number on the screen is the right one would not be fine.

What you will learn. Why checking became the valuable part of knowledge work, and the memo this chapter follows from start to finish.

Why it matters. If AI writes the draft and nobody judges it, nobody did the job.

The first draft got cheap

What changed?

Ana Morales is an accounts payable (AP) analyst at Brightline Wholesale Supply in Columbus, Ohio. AP is the team that pays the company's bills. When Ana joined, new analysts built the vendor reports themselves, line by line. They made mistakes, found them, and slowly learned which numbers looked wrong. Today an AI assistant builds those reports in a minute. They look clean and read well.

The first draft of research, writing and analysis is now cheap, and it arrives fluent (smooth and well written), confident and fast. None of those three things is evidence that it is right. Chapter 0.1 explained why: the model predicts likely text, it has no built-in truth-checker, and it sounds just as sure when it is wrong.

So what is left for people?

Judging the draft:

• Is it true?
• Does it answer the question that matters?
• What is missing?
• What should happen next, and who decides?

That is critical thinking: working out what is true and what is only claimed. The decision you make at the end, usually without certainty, is judgment. A judgment is different from an opinion. An opinion is a reaction: "That feels wrong." A judgment comes with reasons you can state: "Here is my best argument, and here are the facts behind it."

Do employers really pay for this?

Yes, and they now name it. In November 2026, PwC US, a large accounting and consulting firm, announced a Critical Thinking and Judgment certificate for more than 85,000 partners and employees, to be completed by June 2027. Its reason is this lesson's: AI now does the basic tasks that used to teach new staff good judgment, so the firm must teach that judgment directly.¹ The program is practice, not theory. In one practice exercise, a tax professional reviews a tax analysis that AI wrote. The analysis is convincing, but it may be incomplete. The professional must question its sources, consider other explanations, and find what is missing.² That is Ana's task in this chapter, in a different field.

PwC is not alone. In 2026, IBM asked 1,500 chief HR officers which skill the workforce needs most for AI. 71% chose the ability to supervise, check and override AI output. In the same study, about six in ten employees said they worry that AI is weakening their own skills.³ The checkpoint in this chapter is the book's own method, not PwC's.

Can judgment be taught?

Partly. The research gives an answer in two parts.

The method can be taught. A review of many studies found that teaching critical thinking directly does improve it. The gain was real but modest. It was largest when learners discussed real problems with a teacher or mentor.⁴

Judgment has to be built. Critical thinking depends on knowing the subject well. A person who checks a finance memo well does not automatically check a medical claim well.⁵ Judgment also becomes reliable only where results follow regular patterns and you have had the chance to learn them. Feeling confident is not a sign of being right.⁶ Training still helps. In a large forecasting study, people who were trained to think about probabilities made more accurate predictions.⁷

So this chapter gives you the method and a first practice. It does not give you finished judgment. That grows as you use the checkpoint on real work in your own field and see how your decisions turn out. This is why the chapter ends with a task from your own work. PwC's own results will not be known until after June 2027.

What happens when nobody checks?

Two real cases show the cost.

• Invented court cases. In 2023, two lawyers in New York gave a court a document that referred to earlier court cases. ChatGPT had invented those cases. The lawyers did not look them up. One of them had only asked ChatGPT whether the cases were real, and it said yes. The judge ordered the lawyers and their firm to pay a $5,000 penalty.⁸
• A chatbot's wrong promise. In 2024, a man asked an airline's website chatbot about a lower fare for people traveling after a death in the family. The chatbot said he could ask for the lower fare after his flight. The airline's real policy did not allow that. A tribunal in Canada (a kind of court) decided that the airline is responsible for everything on its website, including what the chatbot says. It ordered the airline to pay him about 812 Canadian dollars.⁹

In both cases the AI's answer was well written and confident, and the people who used it did not check it.

The memo this chapter follows

Words you need for this chapter

• Early-payment discount: a vendor takes a little off the bill if you pay early.
• "2/10 net 30": take 2% off if you pay within 10 days. Otherwise pay the full amount within 30 days. "Net 30" alone means no discount.
• Line of credit: money the bank lends on demand. Brightline pays 9% a year on what it borrows.
• Payment run: the day each week when AP pays the bills that are due. Brightline's is Friday.
• Q3: the third quarter of the year, July to September.
• Annualized return: what a discount is worth if you think of it as interest earned over a year. Take 2/10 net 30. You earn 2% for paying 20 days early. A year holds about 18 periods of 20 days, so earning 2% each time comes to roughly 36% a year. The exact figure is about 37%, because the 2% is earned on the 98% you pay. That is far more than the 2% it looks like.

Figure 0.4.1. How to read "2/10 net 30". Paying on day 10 means paying 20 days early to earn 2%.

What happened at Brightline?

Dave Kowalski is Brightline's controller, the manager in charge of its accounting. He typed one request into an AI assistant: "Confirm that we should take every early-payment discount our vendors offer." Seconds later he had this memo. He was about to forward it to the chief financial officer (CFO), who is the company's top finance manager:

Recommendation. Yes. Brightline should take every early-payment discount, starting with the November 6 payment run.

1. The savings are large. In September, Brightline paid $176,100.25 to vendors. A 2% discount on that amount is $3,522.01 a month, or about $42,264 a year.
2. It is standard practice. Most distributors of Brightline's size take every discount they are offered.
3. It costs nothing. Paying 20 days early costs Brightline nothing, because the cash would be paid out anyway.
4. Every vendor benefits. Vendors like being paid early, so relationships will improve across all seven vendors.

It reads well. Not every AI would write this memo: asked the same way, a careful model may add conditions. This one did not. Ana asked Dave for five minutes before he sent it. The rest of this chapter is what she did with them.

Key words

• Critical thinking: working out what is true and what is only claimed.
• Judgment: the decision you make at the end, with reasons, usually without being certain.
• Opinion: a reaction without stated reasons.
• Early-payment discount: money a vendor takes off the bill if you pay early.