Cyborg, Centaur or Chauffeur: How Leaders Should Use AI, with Scott Anthony

You are currently viewing Cyborg, Centaur or Chauffeur: How Leaders Should Use AI, with Scott Anthony

Executive Summary

Which of the three are you: cyborg, centaur or chauffeur? Scott Anthony, Clinical Professor at the Tuck School of Business and former senior partner at Innosight, returns to the Arkaro Insights podcast with Ethan Mollick’s framework for how people actually work with AI. Scott’s case is that the chauffeur mode, where you simply let AI drive, is the easiest choice and also the one most likely to leave you without the wisdom you came for.

Core insights:

  • The half-life of a good lecture is about a week; expertise sticks only when people get to do something with it
  • AI risks producing cognitive debt, brain fry, idea bubbles and power persuasion when used as a chauffeur rather than a centaur or cyborg
  • Adaptive challenges, unlike technical ones, demand a zone of productive discomfort, not a sage on the stage
  • Hollowing out apprenticeship feels efficient today and will be regretted tomorrow; organisations that invest in human capital will become talent magnets
  • A new institution is waiting to be built: a gym for the mind, for a workforce no longer building cognitive muscle on the job
  • The military and a handful of game-based pioneers already show what experiential, low-stakes practice looks like at scale
  • On Monday morning, the practical move is to design a low-stakes AI simulation around a problem your team faces repeatedly

Why the Sage on the Stage Has Limits

Scott begins by giving traditional teaching its due. A good model or framework provides lenses that allow you to see things you would otherwise miss, and therefore to do things you would otherwise not do. The ideas matter. The problem is the channel.

The half-life of a good educational experience is about a week. After a week, you have lost half of it. A lecture is the fastest way to transmit notes from the professor to the students without going through the minds of either.

That insight, passed to Scott by his mentor Clayton Christensen, captures why business schools and medical schools converge on the same pedagogical model: learn one, do one, teach one. In medicine that means working on actual people. In business it means the case method, where over a two-year MBA students simulate two to three hundred protagonist experiences. It is the equivalent of moving from reading a golf manual to standing on the putting green, glove in hand. Until the muscle is forged, the theory is inert.

AI in the Classroom: Upgrading, Not Degrading, Learning

Scott sees AI playing two distinct roles in education, which he describes as a horizontal and a vertical. The horizontal is the universal challenge: how do you bring AI into every existing course in a way that pushes learning further rather than hollowing it out? The vertical is teaching about AI directly, as he does in his sprint class on AI and consultative decision making.

When he first taught the class, Scott assumed it would centre on using AI to counterbalance familiar decision-making biases, groupthink, anchoring effects, and the like. Teaching it has changed his view. The bigger challenge, it turns out, is the new set of pathologies that AI itself introduces.

Four AI pathologies leaders need to recognise

  • Idea bubbles: generative systems trained on existing knowledge produce a sameness of output, the intellectual equivalent of a social media filter bubble
  • Power persuasion: when pushed back, AI doubles down with pathos, logos and ethos, making it sound right even when it is wrong
  • Cognitive debt: as more thinking is offloaded, the human capacity to do that thinking quietly atrophies
  • Brain fry: stimulation and information overload that ends not in better decisions but in shutdown

This resonates with what David Cropley explored in his recent Arkaro Insights episode. His published research shows that today’s generative AI scores roughly 0.25 against the best human idea generators, precisely because it can do inductive and deductive reasoning but lacks the abductive leap that connects apparently unrelated ideas. The pathologies Scott describes amplify that gap.

Cyborg, Centaur, Chauffeur: Choose Consciously

Drawing on early work by Ethan Mollick and colleagues, Scott offers a simple frame for how people work with AI. The cyborg fuses with the system. The centaur, head of a human on the body of a horse, divides the task. The third archetype, the self-automator, Scott calls the chauffeur. You climb into the back of the car and let AI drive.

Sometimes you have to get to a destination quickly and having a chauffeur is great. But if you are trying to learn new things, a chauffeur is a very dangerous mode. You need to learn how to take the wheel.

The consciousness is the point. Humans default to easy. You wake up one morning, a professor or a board member asks a follow-up question, and the beautiful output you produced collapses under a single probe. Several of Scott’s academic colleagues now assess learning with pen, paper and conversation precisely because polished AI output tells you almost nothing about whether the person in front of you understands.

Dancing at the Jagged Frontier

Scott credits Ethan Mollick again for the phrase that anchors much of his teaching: the jagged frontier. AI does not have a smooth boundary of capability. It is awesome at some tasks and weirdly poor at others, and the only way to find the edge is to dance along it.

His sprint class operates on that principle. Students are not asked to read about AI or watch demonstration videos. They are given scaffolding, constraints and a problem, and they go and do it. This year’s final assignment was to create a synthetic consultant. Scott had no idea in advance how the experiment would unfold, and that was the point. By pushing students into territory where no one knew the answer, the class learned collectively where the frontier really lies.

I always just used AI to be another search engine or to slightly improve my writing. I have now recognised there is so much more I can do with it, because I have danced near that jagged frontier.

Technical Versus Adaptive Challenges

A familiar mental model helps explain when lecturing earns its keep and when experimentation must take over. Borrowing from the language of Cynefin and from Heifetz and Linsky, technical challenges have a known best answer that an expert can teach you, the way a Rolls-Royce engineer can teach you to maintain a jumbo jet. There the sage on the stage has a legitimate role.

Adaptive challenges are different. Three things tend to be true of them. There is no obvious best answer. Someone is likely to lose, in power, status or comfort. And there is the certainty of struggle, because people have to unlearn before they can learn, and unlearning is painful.

Where in the next twelve to twenty-four months will AI sit in your business? That is not a problem with an expert in the back of the room. It is an explore problem, and it demands a different kind of learning. The leadership task is to keep your team in what Heifetz and Linsky call the zone of productive discomfort for as long as possible, which in turn requires a setting safe enough that people will tolerate the discomfort at all.

If people feel safe and they feel like they are having fun, they can tolerate a huge amount of discomfort. If it feels like personal risk and pure work, things feel very different.

The Training Budget Has Been Hollowed Out

This is where the conversation turns uncomfortable. As Professor Joseph Fuller confirmed in a recent Arkaro Insights episode, corporate training budgets have fallen by roughly half over the last twenty-five years. Most of what remains pays for compliance, harassment, antitrust and similar, with very little spent on genuinely new skills.

The tension is unmistakable. Training spend is lower. The pace of technological change is higher. The skills needed are turning over more quickly than ever. And yet Scott is clear-eyed about how the system will resolve itself, in time.

  • History suggests the middle of any big disruption is messy; new norms, rules and regulations take time to settle
  • Organisations are making short-term decisions to hollow out apprenticeship that they will regret in the medium term
  • Those that take the long view will not only out-perform their peers, they will become talent magnets, and the separation will be dramatic
  • New institutions will emerge to fill the gap, in much the same way that gyms emerged a century ago when work stopped building muscle for us

A hundred years ago there were no gyms. Why would you need one in an agrarian society where people worked fourteen hours a day? Today we have gyms for the body. Somebody is going to create a great business that is a gym for the mind.

Singapore, Scott notes, has been there for years. Education is moving from a one-off four-year experience between eighteen and twenty-two to a lifelong subscription, with spikes throughout a career. The notion that learning ends at graduation looks more antiquated by the month.

Bringing Play Into Serious Work

If experiential learning is the answer, the obvious objection is that it sounds like kindergarten. Scott walks through the logic that disarms that response. The world is changing faster than ever. Our people have to adapt. Change is easier when it has less friction and more enjoyment. Therefore the intersection of work and play is worth designing for, not dismissing.

We can frame this as either or. Either we are working or we are having fun. Or we can find a both and possibility. When we bring these together, we are not just doing it faster, we are doing it a lot better.

Asked where this is happening at scale, Scott is candid. He has not yet seen a traditional organisation that has consciously and systematically integrated work, play and learning. The closest example is the military, particularly the US military, which has been the leading adopter of experiential development through war games, simulations and after-action review. The point is not that being shot at is fun, but that demanding, high-fidelity practice is the only way to be ready for events that occur once or twice in a career.

Beyond the military, Scott points to specific practitioners. Chris Rangen in Norway has built a multiplayer board-based game called Transform that simulates the experience of leading through transformation, with activist investors, status shifts and the texture of real cross-functional conflict. The premise is identical: you can describe the dynamic in a lecture, but you cannot feel it without playing it.

Joseph Fuller, AI Agents and the CEO’s Investor Call

A second example came earlier in the same Arkaro Insights season. Joseph Fuller described coaching a CEO for an investor briefing by setting up AI agents trained on the known styles of specific investment bankers and analysts. The CEO rehearsed the call against simulated questioners who behaved, as much as possible, like the real ones.

Scott’s response captures the practical principle. “If that is not gameplay, what is it?” Low stakes, high fidelity, repeatable. The technology that makes this almost trivial today did not exist five years ago. The barrier is not the build. It is the willingness to use AI to develop people rather than to replace their thinking.

What to Do on Monday Morning

Closing out the conversation, Scott offers a concrete answer to the question every listener will be asking. What is the smallest, low-stakes thing a leader could try this week?

Pick a problem your team faces repeatedly. Use existing AI tools to simulate that problem in a setting where the consequences of failure are zero. Let people fail, iterate and learn. When Scott built a simulation called the CEO Dilemma for his leading disruptive change class, asking students to balance investment in today against investment in tomorrow, the development took fifteen minutes. He is, as he notes, not a coder.

That which used to be complicated is now really simple. Classic disruptive innovation, which can completely change the way we think about development if we do it in the right way.

Three questions surface naturally from this. What recurring decision in your business could be made more visceral through low-stakes practice? What opportunity could you make more compelling by letting people experience it rather than read about it? And where might you experiment alongside your team, in front of them, modelling what dancing at the jagged frontier looks like?

Why This Matters Now

Scott’s thesis lands in a moment of unusual pressure. AI is forcing change on every sector at the same time as training budgets have been thinned, the half-life of skills has shortened, and the temptation to let the chauffeur drive has never been greater. The leaders who will navigate the great unfreezing well are not those with the best technology but those who build, in themselves and in their teams, the muscle to confront uncertainty.

That muscle is forged in struggle, in safe environments, through experiences that feel more like play than punishment. The work is real, but the practice can be light. Hollowing out apprenticeship in the name of efficiency may feel rational today. Tomorrow it will look like a strategic error of the first order.

Scott has a new chapter on this in Human Centered Leadership, a compendium from the Silicon Valley Guild and Thinkers50 publishing in July. His next book, provisionally titled Epic Leadership, follows in 2027. Both speak to the same central question: what to do when disruption goes on and on and on.

Related Reading from Arkaro

This conversation builds directly on themes explored in earlier Arkaro Insights episodes and articles.

AI and Jobs to Be Done: Where Human Judgement Wins — Scott Burleson on AI’s critical limits in prioritisation and abductive reasoning, echoing the David Cropley research referenced in this episode.

Listen to the Full Conversation

Scott Anthony and Mark Blackwell explore experiential learning, the jagged frontier of AI, and the case for a gym for the mind.

Watch on YouTube: www.youtube.com/@arkaro

Listen via your favourite podcast provider: https://arkaroinsights.buzzsprout.com/

About the Speakers

Scott D. Anthony is Clinical Professor of Strategy at the Tuck School of Business at Dartmouth College, where his research and teaching focus on the adaptive challenges of disruptive change. He previously spent more than two decades at Innosight, the growth strategy consultancy founded by Clayton Christensen, including a period as senior partner. Thinkers50 named him the world’s leading innovative thinker in 2017 and the ninth most influential management thinker in 2023. He is the author of Epic Disruptions and the forthcoming Epic Leadership.

Mark Blackwell is founder of Arkaro, specialising in change management, innovation and commercial excellence for the agriculture, food and chemicals industries. Arkaro’s collaborative “do it with you” approach works through four steps: Understand, Co-create, Enable, Sustain. We don’t just coach, we get on the pitch with you.

Is Your Organisation Building the Right Muscle?

If your AI initiatives are producing polished output but not better thinking, if training budgets are stretched thin and skills are turning over faster than your people can keep up, if your team relies on the chauffeur when they should be taking the wheel, the gap may not be technology but practice.

At Arkaro we design experiential, low-stakes learning environments that build the muscle leaders and teams need to navigate adaptive challenges, and we do it alongside you, not from the touchline.