What work is arranged around today
In a steam-era factory, everything was arranged around the shaft, and no one thought to question it until much later. In the new AI era, organizations are deploying the new tools to automate their existing processes, and thereby automating the same frustration and distrust that were there to begin with.
In our consulting work, we have learned that work is arranged around something less visible: the commitments people make to one another about who will deliver what, to whom, by when and to what standard. That is what a new tool disturbs, often without anyone paying attention to it.
Electricity removed a limit on where machines could stand. AI is removing a limit on how fast work can be produced. When producing the work stops being the slow part, the slow part becomes everything the work rests on: agreeing what the customer needs, judging whether the work meets it, and keeping the commitments people make to one another. Here the parallel with electricity ends. The shaft could be taken out. The network of commitments cannot, because it is how the work gets done. AI puts more weight on it, not less.
We saw this with a team at a large US energy company that we helped start using AI in its software development work. After a few weeks of groundwork, the tools were helping them build what the business had asked for. They also surfaced problems in older code that the team had long since learned to live with. That was the easy part. What came next was harder: a conversation no one wanted to have, about who would raise them and how their manager would hear them. That conversation, not the tool, decided whether anything would change. In our experience, this is where AI's findings most often stall in large organizations: on the way up, not in the technology. Here they didn't. We coached the team on how to raise them, and they went to management that week.
Much of the public debate about AI turns on keeping a "human in the loop." In our experience, that means more than someone checking the output. Whether work is good enough is always someone's call, made against what a particular customer needs and the standards a team shares. Those needs are worked out in conversation; they are not found in the output. Making that call takes judgment that comes from practice: knowing when the AI's work can be used, when it misleads, and when something needs repair. It also takes someone who cares how the work lands, and who keeps the promises made around it. Where no one is doing that, the gains stay small and the exposure, to customers and to the business, grows.
This is also why the payoff is slow. That judgment is a skill, and people build it the way they build any skill: by doing the work, falling short, and learning from those further along. When AI does the work junior staff used to learn on, that path narrows. Where will your next experienced managers come from? That is a workforce and training question as much as a technology one, and the organizations that gain most are likely to be the ones that answer it deliberately.
We are grateful to the philosopher B. Scot Rousse for the clearest statement of this argument we have seen. Rousse studies expertise, skilled work and what AI means for both. In a new chapter in The Oxford Handbook of the Philosophy of Work, he argues that AI succeeds only where people stay in the loop "not only as verifiers of output," but as the ones who care about the results and keep the commitments the business depends on. "Machines emit words," he writes, "but they cannot make promises, sustain trust, or take responsibility." Promises, trust and responsibility stay with your people, and they are what any AI investment rests on.
