Who Teaches Judgment When AI Writes the Draft?

A junior employee used to learn by doing the first pass. They gathered the papers, drafted the note, checked a figure and watched someone more experienced mark what was wrong. A tool can now produce a polished pass in seconds. The open question is not whether that speed is useful. It is whether anyone still teaches the judgment the slow pass used to train.

On 9 November 2026 Business Insider reported a concrete answer from one large employer. The page timestamp is 13:10 UTC that day. PwC US plans a certification in critical thinking and judgment. The program is set to start in December. That is a plan to rebuild practice. It is not evidence that judgment has already been lost, and it is not evidence that the certificate will restore it.

A December start is not a finished certificate

Yolanda Seals-Coffield, chief people and inclusion officer at PwC US, told Business Insider that more than 85,000 employees will take part. The group is people across the firm's US and Mexico offices, plus those in its Acceleration Centers. The article describes those centers as support for client work and business operations.¹

She said that as AI takes on more foundational tasks, some of the repetition that helped people develop critical thinking and judgment will no longer be there. The sentence in the article uses a contraction. The line here is a paraphrase, not a quotation. She also said that first-year associates are already doing different work than they did two years ago. That is her observation. The article does not publish a time-use study behind it.¹

The same interview is careful about hiring. PwC continues to recruit on college campuses and makes offers to the majority of its summer interns. "Majority" is not a percentage. The firm declined to release year-over-year figures on entry-level hiring. She said the firm is hiring more engineers and technologists than ever, across its businesses, and that engineers, already familiar in advisory, are now in tax and assurance as well. Those are statements about recruiting mix. They are not a count of junior roles removed.¹

PwC has identified 15 AI skills and 15 human skills, with empathy and adaptability given as examples of the human set, that it says will set employees apart. This certificate centers on critical thinking and judgment. Seals-Coffield described the aim as a framework that helps people evaluate information, establish context, challenge information and assumptions, refine analysis and communicate recommendations. That is the stated aim. The announcement does not show that the framework already does those things.¹

The coach practices. People still assess.

The company expects employees to finish over roughly six months. The first step is an assessment of thinking style and how it might shape judgment: whether someone questions assumptions, reflects, or is a pragmatist. Those three labels are the examples in the article. They are not a published test with a scoring manual.¹

Seals-Coffield said the firm is working with professors and veteran PwC staff to develop classes. That is work in progress. The article does not name the professors, and it does not present a finished catalogue. The next step is what she called realistic simulations, so people can apply the framework to their work. After that, a personalized AI coach is meant to help them use the lessons on daily tasks.¹

The examples of the coach are possibilities, not a shipped feature list. It could help people practice a difficult conversation or prepare for a client meeting. That might mean uploading a proposal, having the coach review it, and anticipating questions. It could also help identify which team member should take which role. The article says the content will be customized by seniority, and that the firm sees critical thinking as a complement to instruction on responsible use of AI and on professional skepticism.¹

The assessment design matters more than the label "certificate". The program rolls out in stages, so people can apply the lessons before development leaders evaluate proficiency inside the performance-review process. Seals-Coffield said: "The idea of this certification is not to design a traditional pass/fail exam, but to give people these repeatable skills and then to encourage and create the space for them to be assessed against them at year-end." The target is completion by the end of the firm's fiscal year in June 2027. After that, she said, the coach can deliver further instructional content as the company decides updates are needed. "We will continue to assess our people against their AI and human skills proficiencies throughout the course of their career," she said.¹

Read that sequence in order. Classes are still being built. Simulations come next. The coach is described as practice: conversations, a proposal, a suggestion about roles. The year-end judgment sits with development leaders. Nothing in the announcement says the coach scores the certificate, decides promotion, or has been shown to reduce errors on client work. There is no published result here on decision quality, error rates or promotion. An 85,000-person program is also not a template a 30-person organization can copy. The transferable idea is narrower: judgment needs practice, feedback and a human evaluation. Access to a tool, or a seat in a generic AI seminar, is not that evaluation.

One sentence about job ads is not this program

Business Insider adds a line from "a recent PwC analysis of US job listings". Entry-level roles exposed to AI were far more likely to include traditionally senior-level skills, including judgment and leadership, than in 2019. That sentence has no multiple, no sample size and a comparison with 2019. It is not the same claim as a seven-times gap between the most exposed junior jobs and the least exposed ones.¹

PwC Cyprus published a press page for the 2026 AI Jobs Barometer that does use the stronger wording. A bullet says AI-exposed US entry-level roles are seven times more likely to require traditionally senior-level skills such as judgement and leadership, that these roles grew 35% since 2019, and that other entry-level roles declined by 10%. The body says the entry-level analysis covers 2.4 million US jobs, and that the most exposed entry-level roles are seven times more likely to require human-intensive skills such as leadership, creativity or face-to-face interactions. The same page is not a methods document. A bullet prints headcount growth of 53% versus 36%. A later paragraph prints 52% relative to 36%. A bullet prints the wider jobs market at 9%. A later paragraph prints 8.6%. This article does not use those productivity or wage figures as findings. The mismatch is a reason to treat the page as a local press restatement.²

The limits on "seven times", on the 35% and on the 10% are already set out in why junior advertisements now ask for senior skills. That reading used the global release and the report, including the US chart. This certificate does not test that study, and that study does not test this certificate. A posting that asks for judgment is not proof a new hire has it. A training plan is not proof the posting was caused by AI.

A different survey speaks to hiring difficulty, not to PwC's classroom. On 29 October 2026 Randstad said 61% of employers report that finding talent with communication and stakeholder management skills is harder than sourcing technical skills alone. The body of the same page uses a wider phrase for that 61%: human capabilities such as critical thinking and communication. The primary survey was 1,100 interviews with senior business and technology decision-makers, 100 in each of 11 markets, in the first half of August 2026. It is not a study of whether a certificate changes client work. The fuller reading is in why AI hiring needs judgment, not only code.³

Going past one's training is a third question. See why AI letting apprentices work beyond their training is not mastery. Who receives any AI training at all is a fourth. See why AI training reaches bosses more often than beginners.

What a smaller organization can still teach

The steps below are an editorial suggestion. PwC did not test them.

Start with one recurring decision, not a slogan about skills. A nonprofit might use grant-eligibility reviews. A small manufacturer might use exceptions in supplier quality. A department might use the escalation of a customer complaint. Name the information the decision needs, the errors that matter, and the person who remains accountable. The scene is an illustration. It is not a PwC case.

Then run three versions of the same short case. In the first, the person works without the tool, so there is a baseline of their reasoning. In the second, they use the tool and must mark sources, uncertainty and the alternatives they rejected. In the third, they receive a deliberately flawed generated answer and must say what is wrong, what evidence is missing, and whether the case should be escalated. These versions are proposed exercises. They are not findings from the PwC program.

Keep a short decision record: what was recommended, which evidence supports it, which assumption would change the outcome, and who approved the action. A manager should review the reasoning, not only the speed or the polish. Ten minutes on one difficult case can teach more than an hour of generic prompting advice. That, too, is a suggestion, not a measured result.

If you track learning, use criteria you can repeat: whether the person spots an error, the quality of the evidence, whether they escalate when they should, whether the explanation is clear, and whether the judgment is consistent on cases of similar difficulty. Do not treat self-reported confidence, the number of prompts, or a course completion as proof of better judgment. For a high-stakes decision, keep a qualified person in the loop and keep the privacy safeguards you already owe.

A counterargument is fair. A tool can be a patient tutor. It can generate cases, explain an unfamiliar idea and answer at once. For someone who never had a good supervisor, that may widen access to practice. Fluent text is not the same as a correct text, and the tool does not carry the legal, ethical or organizational accountability for the decision. A credible exercise teaches when to challenge the draft and when to call a human expert.

The legal line is narrower than the certificate

This article is not legal advice. A critical-thinking certificate is not, by itself, compliance with the EU AI Act.

Article 4, in the text the Commission service desk shows as amended, says providers and deployers shall take measures to support the AI literacy of their staff and of other people who operate AI systems on their behalf. They take account of technical knowledge, experience, education and training, the context of use, and the people or groups on whom the systems are used. The same paragraph says this obligation does not require anyone to guarantee a specific level of AI literacy for any individual. The service desk says the wording reflects the EUR-Lex consolidated text as at 27 July 2026, including Digital Omnibus amendments. The summary on that page is not legally binding. A judgment course may support literacy. Completing it is not proof the obligation is met, and the article does not require this kind of certificate.⁴

Article 5(1)(f) prohibits placing on the market, putting into service for this specific purpose, or using AI systems to infer emotions of a natural person in the workplace and in education institutions, except where the use is intended for medical or safety reasons. The PwC announcement describes a thinking-style assessment: questions assumptions, reflects, or is a pragmatist. It does not describe inference of emotions. This article does not treat that assessment as a prohibited emotion-recognition system on the text of the announcement alone. The service desk says prohibited practices are enforceable from 2 August 2026.⁵⁷

Annex III, point 4(b), lists AI systems intended to make decisions affecting the terms of a work relationship, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate performance and behaviour in such a relationship. The announcement says development leaders evaluate proficiency in the performance review, and it describes the coach as practice support, including a possible suggestion about who should take which role. That description, alone, is not a finding that the coach is a system intended to evaluate performance or to allocate tasks. Classification depends on the intended purpose. The service desk says Annex III rules apply from 2 December 2027.⁶⁷

Article 26 is conditional on a system that is high-risk. Paragraph 7 says that before putting a high-risk AI system into service or using it at the workplace, employers shall inform workers' representatives and the affected workers. Paragraph 11 says that, where Annex III high-risk systems make or assist decisions related to natural persons, deployers shall inform those persons. The announcement does not classify the coach. These duties are not a finding about PwC. They are the questions to ask if a later tool is intended to score people for promotion or to assign work from personal traits.⁸

The announcement puts the year-end judgment with development leaders, not with the coach. This article does not analyse whether a later scoring tool would be a decision taken by the tool alone under data-protection law. The official GDPR text did not open in this pass.

How hilighter can help

hilighter does not run the PwC certificate, and it does not score critical thinking. It can make the practice under a role visible. Teams can map skills by role, compare people with those requirements in skills and talent insights, record proof of a correction or a decision in skills evidence and recognition, and set personalized growth paths. A manager's review of that evidence sits in manager assessments. The AI job impact view is indicative. It is not a prediction that a junior role will disappear, and it is not a PwC or Randstad readout. The Free plan covers up to three job roles. The number of employees is not capped.

Sources

1. Tim Paradis, Business Insider, PwC plans to certify workers in critical thinking as AI changes their jobs, page timestamp 9 November 2026, 13:10 UTC. Yolanda Seals-Coffield, chief people and inclusion officer, PwC US. More than 85,000 employees across US and Mexico offices and Acceleration Centers. Program set to start in December. Roughly six months. Thinking-style assessment. Classes being developed with professors and veteran staff. Realistic simulations. Personalized AI coach as practice. Year-end assessment in the performance review, not a traditional pass/fail exam. Target: end of fiscal year, June 2027. Campus recruiting continues. Majority of summer interns receive offers. No year-over-year entry-level hiring figures. 15 AI skills and 15 human skills. Job-ad sentence: far more likely than in 2019, no multiple.
2. PwC Cyprus, 2026 AI Jobs Barometer press page. Bullet: seven times, judgement and leadership, +35% since 2019, other entry-level roles -10%. Body: 2.4 million US entry-level jobs; seven times versus human-intensive skills such as leadership, creativity or face-to-face interactions. The same page prints 53% versus 36% in a bullet and 52% relative to 36% in a later paragraph, and 9% versus 8.6% for the wider jobs market. Local press restatement, not the methods report.
3. Randstad, 86% of employers are growing tech teams as AI reshapes demand, 29 October 2026. 61% in the bullet (communication and stakeholder management) and 61% in the body (human capabilities such as critical thinking and communication). Primary survey: first half of August 2026, 11 markets, 1,100 interviews, 100 per market, senior decision-makers. Evalueserve used for sampling.
4. European Commission, AI Act Service Desk, Article 4: AI literacy. Wording reflects the EUR-Lex consolidated version as at 27 July 2026, with Digital Omnibus amendments. The summary is not legally binding. Paragraph 1 does not require a guaranteed level of literacy for any individual.
5. European Commission, AI Act Service Desk, Article 5: Prohibited AI practices, paragraph 1(f): emotion inference in the workplace and in education institutions, with a medical or safety exception.
6. European Commission, AI Act Service Desk, Annex III, point 4(b): decisions on terms of work, promotion or termination; task allocation based on individual behaviour or personal traits; monitoring and evaluating performance and behaviour.
7. European Commission, AI Act Service Desk, When does enforcement start?. Prohibited practices enforceable from 2 August 2026. Annex III rules from 2 December 2027.
8. European Commission, AI Act Service Desk, Article 26, paragraphs 7 and 11. Information duties for employers and for people subject to Annex III high-risk systems that make or assist decisions. Conditional on a high-risk system.

Sources last reviewed: 12 November 2026. The Business Insider page is dated 9 November 2026 in its timestamp, not 10 November. The repetition sentence is paraphrased because the published line uses a contraction. The coach examples are the article's "could" and "might", not a product sheet. The Cyprus page disagrees with itself on figures this article does not adopt. The seven-times limits live in the 10 November hilighter article, which used the global report. EUR-Lex did not open for the GDPR text (403), so Article 22 is not quoted. EU dates follow the Commission service desk and are not legal advice. Neither an AI-detector score nor the absence of Google penalties can be guaranteed.