How AI is reshaping thinking, teaching, and judgment at work

Executive overview

Outsourcing thinking to AI feels effortless — but it quietly erodes the judgment that makes knowledge workers valuable. Professor Scott Anthony draws a hard line: AI for brainstorming and research, never for first drafts of anything that requires original thought.

The risk compounds across generations. Students who never struggle with blank pages may never develop the critical reasoning that experience is supposed to build.

The core insight: augment with AI, never replace your thinking — because once you stop, it's very hard to start again.

Where AI helps and where it doesn't

• Brainstorming and shortlisting (e.g. which disruptions to include in a book) — highly effective
• Titling, theme extraction, and wrapping structure around original content — useful
• Research partnership: testing whether an idea has precedent, surfacing adjacent literature
• First drafts of anything requiring original reasoning — actively harmful to skill development
• Panel prep for low-stakes facilitation — fine to outsource; the output is raw material, not the work

Context engineering and multi-model prompting

• Set a persona or stance (skeptic, supporter) before prompting to get genuinely different perspectives
• Run the same prompt in multiple models (Claude, ChatGPT, Gemini) simultaneously
• Play models against each other: share one model's response with another and ask for a reaction
• End complex prompts with "ask me questions so you can be 95% confident" — the questions reveal assumptions
• Models converge on fundamentals but diverge early; the divergence is where the value is

The critical thinking risk

• AI offers the path of least resistance; humans are wired to take it
• Getting a first draft from AI feels different from buying an answer key — but the cognitive effect is similar
• Early adopters risk hollowing out their capabilities: short-term efficiency, long-term loss of judgment
• The tell: LinkedIn thinkers whose posts no longer sound like them
• Protecting judgment requires active willpower — sitting in discomfort is the work

How Scott teaches in the AI era

• Courses focus on uncertainty, hearing dissenting voices, and acting without complete data — hard to outsource
• Laptops closed, phones off in class; 50% of grade is in-class discussion quality
• Students may use AI to prepare, but must demonstrate live reasoning in the room
• New final assignment option: design an AI tool to teach a concept to a real leader — and show all the prompts and back-and-forth
• Showing the work matters more than the polished output; elegant results with no visible process earn low marks
• Piloting 30-minute end-of-term interviews to test whether learning actually happened

Consulting and the Gutenberg parallel

• The Catholic Church funded the printing press; it later enabled the Reformation and undermined the Church
• Large consultancies are in the same position: AI boosts efficiency now, but clients will eventually need them less
• A second risk: the generation doing junior analytical work today won't be replaced — so the next generation of senior judgment-holders won't exist
• Pyramid model under pressure; possible shift to diamond or inverted pyramid (more senior advisors, fewer junior analysts)
• Small firms can now look like large ones — a genuine competitive shift

Recruiting and assessing thinking

• Physical, in-person conversation remains the best filter
• Give candidates an open choice (e.g. pick any one of 10 disruptions, form a view) then push them in the room
• What you're looking for: how they approach an issue, not whether they have the right answer
• Psychometric tests and any standardised format can now be gamed

Judgment, mental models, and adaptive capacity

• Experience builds judgment but also creates blind spots — defaulting to familiar models when context has changed
• Clay Christensen's approach: "I have no opinions. The model has a viewpoint." Anomalies are data to upgrade the model
• A wise leader holds a range of tools and knows when each applies
• Scott's framework in development — adaptive capacity — requires three things:
  • Metacognition: thinking about how you think; holding multiple models
  • Equanimity: ability to pause before being overwhelmed
  • Paradox mindset: turning perceived tensions into possibilities

Five-year planning and the "can I watch" habit

• Inspired by Cal Newport's Slow Productivity: set three professional goals on a five-year horizon, not quarterly
• Scott's current five-year goals (set mid-2024): crack the adaptive capacity framework; publish a peer-reviewed article; win a teaching award
• Longer timelines reduce anxiety and allow work to move at its natural pace
• "Can I watch" ritual: once per five-week teaching cycle, observe a colleague's class without caring about the content
• Watching for: how they open, use physicality, hold attention, structure discussion
• Key lessons: a single well-crafted opening question can drive an hour of rich discussion; great teaching has a performance element — rehearsed presence, not scripted words