Both supporters and detractors of AI seem to agree on one thing - AI can do a lot of work. There is a lot of debate still about whether AI is ready to actually replace workers, and both sides of this debate could well be right about their position. But, irrespective of their positions, haven’t both sides already agreed on what will happen next - that AI will replace workers even if it may never be ready to do so? History suggests that they have. That removing the worker from the work, at least in companies, has long been established and agreed upon as a legitimate, even desirable, goal.

When it comes to work, you think “reconciliation”, “board deck”, “ad campaign”, not “Patelian”, “Nguyenesque”, or “in the style of Garcia”. We seem to have made an active choice to disconnect work from its history and authors, and have decided that uniqueness is not a feature of it. This despite the fact that even something as taken-for-granted as milling - a key piece of most manufacturing - results in a unique output in every instance (a unique microscopic “trace”).

One ironic exception here is that one of the early champions of separating the work from the worker - Frederick Winslow Taylor - is one of the few acknowledged authors with his approach of scientific management being named “Taylorism”. This disembodiment may have pushed us to quiet that voice inside that seems to realize that there is something off about work done by a machine, this intuition that important things about work are irreducibly human.

Human beings, and all life, has a limited lifespan and a limited physicality. Our whole way of being is centered around the fact that we have limited time and limited resources. Every action has to start with an intention, and fit within the resources available. So picking, choosing, and abandoning, i.e., discretion, is fundamentally a part of us. There are some actions where we can articulate our discretion semi-accurately but, by and large, it materializes as automated or sub-conscious thoughts or moves that cannot be put into words. 

Let us go back to milling here for a moment: an expert miller feels feedback from the metal block they are working on through their tool, whether the tool is simple or a machine. In a mostly inarticulable way, this feedback tells them of an emerging imbalance in the work that leads them to make an ever-so-slight push with their palm to offset it. A very determined documenter can build a low fidelity articulation of some of the actions that go into a given piece of work, and most of what makes the worker do the things they do is unarticulated, potentially mostly hard-coded. AI had only this low-fidelity dataset to learn to simulate our discretion from. So even if we marvel at the sophistication of this simulation, it should fundamentally just be that - a simulation. 

The confidence that a simulation of work, however sophisticated, can replace workers must have some basis in the past, as all confidence does. Every big milestone in the history of automation and scaling seems to follow a similar pattern: an ambitious proposal - often precipitated by new technology - for increased decoupling of design and execution, a push back focused on minimizing disruption to the specific workers who are in the previous paradigm, regulation aimed at achieving some but not all minimization, societal compromise that lets some of the automation through, and then a gradual understanding that even the compromised version of the automation was not as powerful as originally thought. 

Taylor’s scientific management proposed taking the knowledge of a job out of the craftsperson’s hands and putting it in management’s. Workers fought its stopwatches, speedups, and job losses. Congress limited the system in federal arsenals, but a milder version spread through industry. As commercial aviation automated the cockpit, the prospect of pilotless flight kept returning. Pilots pushed back against smaller crews and raised safety concerns; regulators kept two pilots on the flight deck while allowing automation to do more of the flying. In the 1980s, GM pursued the “lights-out factory.” Unions bargained over layoffs, retraining, and safety, and factories brought in robots piecemeal. The workerless plant never arrived. 

The pattern itself may be fundamentally sound in many respects, and may have led us to important compromises that preserved genuine benefits. But the pattern has never included organized push back on the framing that technology advances can entirely replace workers. On that front, there always seemed to have been a tacit agreement that there was nothing wrong with the framing itself, only with the execution and timing. 

This tacit agreement may be visibly present in each of our individual histories too. My own framing of school was that it was a place where I would learn existing things that are there to be learned. I joined McKinsey eager to learn and apply a proven way to problem solve and influence companies. There is always a tangible thing to pick up, make a copy of, and add copies of other things to it to make a bigger thing. Though technically not inaccurate, it is a framing that always viewed work as being independent of people that did it. In reality, if I let myself dig under all the calcified assumptions, it should have been clear that the most impactful professionals around me (not always the same as the ones who were recognized as impactful) did their work as a subtle dance between established traditions and immediate reality. 

So why would the potential replacees be so eager to tacitly facilitate, or at least not question, the fantasy of replacement? It might just be because they hold a very important common premise with the replacers. Like our physical world relies on the sun for order (days, years, seasons, climate), our work world seems to rely on an idea of progress-as-an-end-in-itself for direction. Under that idea, we all work for progress and not to create, so something that drives general work forward faster and in a higher volume - automation - is naturally immune to fundamental critique. Any opposition to it tends to instead be focused on negotiating its pace, its disruptions, the distribution of its benefits, and who will be protected along the way.  

Without the belief or at least a concession that work is irreducibly human (and maybe humans need to be engaged in purposeful work as a necessity to live), there is no compelling reason to see augmentation, as opposed to automation, as the natural vehicle for progress. If work itself, and not a human, is at the centre, what is there to augment?