The master argument - and the spine of the third book. We inherited a definition of productivity from the Industrial Revolution: output divided by input, forged in the fires of nineteenth-century factory thinking. It was a useful equation for a world of physical production. It is a catastrophic one for a world of knowledge work and AI-enabled organisations.

Efficiency asks: how do we make the same thing faster and cheaper? Effectiveness asks: are we making the right thing at all?

Most organisations are using AI to become more efficient - to automate what they already do, faster and for less. That is the Victorian trap: applying the most powerful technology in human history to problems that were already outdated before computers existed.

The shift I argue for is not incremental. It is a reorientation of the entire question. Stop asking 'how do we do this more efficiently?' Start asking 'what outcomes do we actually want, and how do we design work backwards from those outcomes?'

This is what the post-AI age demands. Not faster Victorians. Something fundamentally different.

My most original framework - and the one I'm still surprised isn't more widely discussed.

Here is the story. A team at Microsoft worked with a hospital trust to apply machine vision to brain tumour analysis. A process that previously took a consultant oncologist five hours per patient was compressed to thirty minutes - with greater accuracy. The productivity gain was extraordinary.

But the oncologist didn't process an additional nine patients in the time saved. She processed the same number of patients she'd always seen. And she used the four and a half hours she got back to do something machines can't do: she sat with each patient and talked to them. Not about the tumour. About what they were going to do next. About how to tell their family. About what mattered.

That was not a productivity failure. That was the right choice. And the lesson is simple: the measure of AI success is not what you automated. It is what you chose to do with the time it gave back. Every organisation will face this choice. Most are going to waste it on doing more of what they were already doing. The ones that don't will be defined by the choice they made instead.

As machines take on more cognitive work, three human attributes become not just valuable but essential: Creativity, Empathy, and Accountability. These aren't soft skills. They are the hard edge of what will differentiate human contribution in an AI-augmented world - and they're attributes that machines are not going to credibly replicate for decades, if ever.

Creativity - because the problems ahead require thinking we have never had to do before. The playbook doesn't exist. Someone has to imagine it.

Empathy - because in a world of cold algorithmic logic, the ability to understand another human being's emotional reality becomes a genuine strategic differentiator. The oncologist story is the empathy story. So is every moment when a leader needs to read a room rather than a report.

Accountability - because just because the algorithm gives you an answer doesn't make it right. Someone has to own the decision. Someone has to ask whether the output matches what we actually value. That cannot be delegated to a model.

What worries me is that most organisations are moving in exactly the wrong direction: cutting the roles and training budgets that develop these skills, while investing heavily in the tools that make them seem less necessary. The irony is that the better the tools get, the more those three skills will matter.

The word 'robot' comes from a 1920 Czech play - R.U.R. by Karel Čapek. His brother Josef coined the word itself: a shortening of 'robota', the Czech for 'serf' or 'forced labour'. The entire concept of artificial workers, from the moment it entered language, was framed as enslavement.

That framing - machines as servants, tools, or threats - has shaped how we design, deploy, and think about automated systems for a century. We build machines to do what we tell them, completely, without judgment, at scale. And then we're surprised when they replicate our biases, miss the nuance, and produce outputs that are technically correct and humanly wrong.

The opportunity I've been arguing for - most recently in The Good Robot Report and as a central thread in the book-in-progress - is a fundamentally different relationship. Not master and servant. Not operator and tool. Colleague and companion. A machine that helps you be more of what you're trying to be, rather than simply executing what you instruct.

This is not science fiction. It is the direction the most interesting human-machine research is already pointing. And it is the reframe that every AI strategy conversation should begin with, before anyone talks about which platform to buy.

In the Netherlands, they have a name for the paths pedestrians create by cutting across grass rather than using the designed walkways: elephant paths.

Architects call them something more diplomatic: desire paths. Whatever you call them, they are telling you something: the people your design was meant to serve have told you, with their feet, that you got it wrong.

Most organisations design from the inside out. They begin with their systems, their processes, their organisational structure, and they build experiences that serve those structures. Then they wonder why customers find them frustrating, why employees create workarounds, why digital transformation programmes stall halfway through.

Outside-In Thinking reverses the question. Start with the outcome your customer or employee actually needs. Start with the moment of genuine value. Then work backwards to design the systems that serve that moment - rather than starting with the systems and hoping value emerges at the end.

In an AI age, this matters more than ever: the temptation is to automate what you already do. But if what you already do is a system designed for your processes rather than your people's needs, you've just made a faster version of the wrong thing. Outside-In Thinking is how you avoid that trap.