AI Is The Wrong Unit of Analysis

Written in collaboration with Claude, Anthropic's AI assistant - because the argument demands it.

Dario Amodei, CEO of Anthropic, the company that builds the AI I’m writing this with, recently framed this moment as “a test of whether humanity is mature enough to handle what it’s creating.”

He wasn’t warning about the technology. He was warning about us.

I can’t stop thinking about that line, not because it’s alarming, though it is, but because it’s asking a human question, not a technical one. Maturity isn’t something you acquire by upgrading your software. It isn’t a capability you can buy, deploy or prompt-engineer your way into. It’s the accumulated result of learning to hold uncertainty, to stay in relationship under pressure, to be changed by experience rather than merely informed by it.

And yet the entire conversation about AI leadership is structured around the wrong question.

Boardrooms, business schools, executive coaching programmes (including some very good ones) are asking “how do we become AI-ready?” They’re hiring Chief AI Officers, running digital literacy programmes, sending leaders on prompt engineering workshops. They’re treating an adaptive challenge as a technical problem, which Ron Heifetz identified decades ago as the central failure of leadership: reaching for a known answer in the system because the alternative, sitting with genuine uncertainty, is simply too threatening.

I’ve spent thirty years watching this pattern repeat. SAAP implementations. ERP systems. Agile transformation. Digital disruption. Every wave of powerful new technology exposes the same underlying deficit, and every time we respond by focusing on the tool rather than the human infrastructure required to use it wisely. The tool changes. The gap persists.

AI doesn’t create that problem - it reveals it, accelerates it, and, if Amodei is right about the trajectory, makes it impossible to keep ignoring.

* * *

The left-hemisphere trap

Iain McGilchrist’s work on the divided brain offers something precise here that most leadership conversations miss entirely. The left hemisphere of the brain - brilliant, systematic, certain - doesn’t just prefer clarity. It actively suppresses what it can’t categorise. It mistakes the map for the territory. It takes a simplified model of reality and treats it as reality itself.

Our dominant organisational model is essentially a left-hemisphere machine. Hierarchies, KPIs, governance frameworks, planning cycles, reporting structures - all of it is exquisitely designed for categorisation, control and prediction. It excels at what Dave Snowden’s Cynefin calls complicated problems: the kind where there is a known answer somewhere in the system, where expertise applies, where cause and effect are visible.

But the problems AI is being deployed into - infrastructure delivery, healthcare transformation, strategic change at scale - are complex, not complicated. Cause and effect are only visible in retrospect. No individual however credentialled, can lead their way through that from the top.

So here is what happens when you introduce AI into a left-hemisphere organisation: the organisation grabs it and uses it to generate more certainty, faster. More data. More dashboards. More confident decisions at speed.

Which in a genuinely complex system doesn’t produce better outcomes. It produces faster mistakes made with greater conviction.

* * *

The immune system always wins

There’s a deeper reason this pattern is so hard to break, and it has nothing to do with intelligence or intention.

Bob Kegan and Lisa Lahey’s research on immunity to change shows that even when people receive life-threatening medical diagnoses and are told exactly what behavioural changes would save them, fewer than 20% make those changes. Not because they don’t understand. Not because they don’t care. But because they have competing commitments - deeply held, often unconscious assumptions about what keeping themselves safe requires - that are more powerful than any stated intention to change. The immune system rejects the new heart even when the patient needs it to survive.

In organisational cultures this dynamic runs even deeper. Fiona Robertson captures it exactly: “We behave our way to belonging.” In a strong Clan culture - and most large organisations are stronger Clan cultures than they would care to admit - the rules of belonging are more powerful than any change programme. Leaders aren’t resisting AI transformation irrationally. They’re protecting something real: their status, their certainty, their sense of being the person with the answers.

David Rock’s SCARF model shows precisely why AI-era leadership triggers this response so acutely. Status is threatened when you don’t have the answers. Certainty collapses in genuinely complex environments. Autonomy is undermined by algorithmic systems you don’t fully understand. Relatedness frays in distributed, hybrid working. Fairness feels violated when AI-driven decisions are opaque. All five threat responses activate simultaneously. The workshop happens. The credential is acquired. And the culture quietly protects itself back to equilibrium.

This is why the AI leadership gap isn’t a training problem. It’s an immunity to change problem. And until we treat it as such - surfacing competing commitments, working with the belonging dynamic, regulating adaptive pressure rather than eliminating it - we will keep producing leaders who are credentialled but not capable.

* * *

The gap we’re not measuring

What organisations actually need to navigate AI-era complexity has a name. Professor Ruth Crick’s decades of research identifies what she calls Learning Power - eight dispositional capacities that determine whether individuals and organisations can engage productively with uncertainty, novelty and genuine challenge:

Mindful Agency: acting purposefully while remaining adaptive. Sense Making: constructing shared meaning from ambiguous, conflicting information across boundaries. Curiosity: staying in inquiry rather than rushing to closure. Creativity: generating new possibilities rather than choosing between existing options. Hope and Optimism: the resilience to keep engaging when outcomes are uncertain and timescales are long. Belonging: a felt sense of being part of something larger than your employer. Collaboration: working across difference, co-generating rather than merely coordinating. Orientation to Learning: the willingness to be changed by the process itself.

These are measurable. They are developable. And they are almost entirely absent from every AI leadership framework currently on the market.

Instead, those frameworks are built around individual skill acquisition - a complicated-domain response to a complex-domain challenge. The mismatch between the capability being built and the environment it’s being built for is where organisations will fail. And they won’t even know why, because they’ll be measuring the wrong things.

We are spending billions on AI capability and almost nothing on the human capability to use it wisely. That is not a talent gap. It is a systems risk - and it is entirely measurable if you know what to look for.

* * *

Maturity, revisited

Matt Shumer, an AI founder writing recently about the pace of change, described the current moment as being in the “this seems overblown” phase of something much bigger than most people are prepared for. He’s right about the acceleration. He’s right about the urgency. His advice - use the tools seriously, adapt fast, build the habit of learning - is sound as far as it goes.

But his entire framework for what to do is still structured around the individual and the technical. Learn more. Prompt better. Get ahead of the curve. It is, in Heifetz’s terms, a technical response to an adaptive challenge - and a familiar one.

What neither Shumer nor most of the AI leadership conversation have quite got to is this: the question was never whether the technology works. It’s whether the human systems around it are capable of using it wisely. That question doesn’t get answered by better prompts or faster adoption. It gets answered by the slow, difficult, unsexy work of building genuine human infrastructure - the Learning Power, the relational capacity, the adaptive leadership capability that allows organisations to navigate complexity rather than just accelerate through it.

Amodei calls this a test of maturity. He’s right. But maturity, in this sense, is not a disposition. It is a capability. It is built, deliberately, over time, through exactly the kinds of experiences and conditions that our left-hemisphere organisations are currently designed to prevent.

The AI conversation has been asking the wrong question. The right question is not “are we AI-ready?”

It is: do we have the human infrastructure to be ready for anything?

And then, honestly: do we even know how to measure that?

* * *

Victoria Ferrier is a strategic advisor working at the intersection of human capability, complex systems and organisational performance. She is collaborating with Professor Ruth Crick on the application of Learning Power to large-scale infrastructure delivery. This piece was developed in active collaboration with Claude - Anthropic’s AI assistant - as a demonstration of the kind of human-AI partnership the argument advocates for.

References & Sources

Amodei, D. (2025). Machines of Loving Grace. Anthropic. [The essay underpinning the maturity warning that opens this piece]

Shumer, M. (2026). Something Big Is Happening. shumer.dev. [The piece that prompted this response - right about the urgency, asking the wrong question about what to do]

McGilchrist, I. (2009). The Master and His Emissary: The Divided Brain and the Making of the Western World. Yale University Press.

Heifetz, R. & Linsky, M. (2002). Leadership on the Line: Staying Alive through the Dangers of Change. Harvard Business Press.

Kegan, R. & Lahey, L. (2009). Immunity to Change. Harvard Business Press.

Robertson, F. (2016). The Rules of Belonging. [Source of “we behave our way to belonging”]

Rock, D. (2008). SCARF: A Brain-Based Model for Collaborating With and Influencing Others. NeuroLeadership Journal.

Snowden, D. & Boone, M. (2007). A Leader’s Framework for Decision Making. Harvard Business Review. [The Cynefin Framework]

Crick, R.D. et al. (2004). Characterising Learning Power. University of Bristol. [The foundational research behind the eight Learning Power dimensions and the foundation of Professor Crick’s WILD framework for building human infrastructure]