Only Coal and Horses
When I half glanced at a headline and saw ‘horse’ and ‘coal’, my mind decided to compute that as ‘Only Fools and Horses’. After reading the article, I did wonder, what would Del Boy and Rodney make of AI?
Del would be halfway through a pitch on Peckham market for something “AI-powered” that he barely understands but is convinced will sell, leaning into the opportunity. Rodney would be a bit more hesitant, probably asking what happens when the thing works properly and starts to change the game they’re playing in. One leaning into the upside, the other clocking the risk. That tension feels familiar, and it’s probably why the whole “coal or horse” idea has landed as well as it has.
If you’ve come across it, the premise is straightforward. Horses were replaced when machines took over physical labour. Coal became more valuable because it powered the system that replaced them. So, when people look at AI, the instinct is to place themselves somewhere on that spectrum and work out which side they would rather be on. Annie Lowrey’s piece in The Atlantic lays that out clearly, and it’s worth a read. There’s also a follow-on argument from Nicole Williams on PM Researcher that leans into systems thinking and the idea that people working in complex, less predictable environments are better positioned as things shift. Again, there’s something in that, and its helpful context if you want to see how others are interpreting the same trend.
What I keep coming back to though is how this will realistically play out inside organisations, because that tends to be where the neatness of the metaphor starts to unravel a bit.
Some of the numbers people reference give a sense of scale, although they often get read in a more dramatic way than they were intended. Goldman Sachs estimated that generative AI could expose the equivalent of 300 million full-time jobs globally to automation, with around two-thirds of jobs in the US and Europe affected in some way. McKinsey has suggested that activities accounting for up to 30% of hours worked today could be automated by 2030. Those are big numbers, but they are describing exposure to task-level change, not a simple removal of entire roles. They point to how much of what we do is made up of work that can be accelerated or reshaped.
Making that distinction is important because most jobs don’t disappear in one clean movement. They evolve, and they tend to evolve in ways that feel gradual while you’re in them. The more repeatable elements start to fall away first, then the more structured parts, and what remains is the layer that relies on judgement, context, and experience. If you were designing this from first principles, you’d expect that to increase the value of the role, because that’s the part that’s hardest to automate. My experience is that, inside most organisations, it doesn’t always work like that.
Roles tend to get simplified around what can be clearly defined and measured. Outputs, timelines, volume, delivery metrics. The thinking layer, the interpretation, the trade-offs, the judgement calls that sit behind the work, often don’t get captured in the same way.
Given enough time, that creates a slightly odd situation where the work is genuinely complex, but the role looks relatively mechanical from the outside. Once something is framed like a process, it becomes easier to standardise, optimise, or partially automate. AI doesn’t need to replace everything for that to have an impact. By taking on enough of the visible work, the balance of value will start shifting.
You can see traces of that in different parts of professional work already. Programme roles that revolve heavily around reporting rather than shaping delivery. Consulting engagements that place more emphasis on producing outputs quickly than on understanding the problem being solved. Analysis that is measured in terms of volume rather than in terms of what changes because of it. None of these areas have become simple, yet the way they are framed doesn’t always reflect that.
That’s where the systems thinking argument feels slightly incomplete on its own. The idea that operating in complex environments is more durable holds up, and there are good examples of that. Software engineering demand has held up despite advances in AI-assisted coding, in part because organisations are still working out how to apply those tools in real environments. Improvements in medical imaging have led to more scans rather than fewer, which in turn increases the need for interpretation. The underlying work becomes more complex as capability expands.
Even so, you can be operating in that kind of environment and still find your role treated as a delivery function if it is positioned narrowly enough. That is the part that tends to get missed.
When you look at the roles that seem to retain their value, there is a pattern that sits slightly outside the coal versus horse framing. It is less about the category of work and more about where the person sits in relation to decisions. The closer someone is to deciding what happens next, or to carrying accountability for an outcome, the harder it becomes to separate them from the value being created. Where someone is contributing inputs into a decision that sits elsewhere, it becomes easier to compress or substitute parts of that contribution over time.
Instead of trying to work out whether you fit into a category, it becomes more useful to think about where you sit in the flow of decisions. Not where you contribute or support, but where you are expected to make a call or stand behind one. For most people, whilst the answer may not always be comfortable, it is fairly clear when they take a step back and think about it.
Where there is distance from that point, the opportunity tends to sit in how the role is approached rather than in changing it entirely. Getting involved earlier when problems are still being shaped rather than handed over fully defined. Spending more time on framing what success looks like rather than focusing only on delivery. Making the reasoning behind decisions more visible so the value of the thinking is clearer rather than sitting quietly behind the output. Small shifts in behaviour that tend to compound over time.
It also brings into focus how roles evolve if they are left unchecked. Work that drifts towards reporting, production, or process handling tends to reinforce a particular perception of value. Once that perception settles, it becomes harder to reposition the role around judgement or decision-making.
Another area that becomes more relevant in this context, although it is not always framed this way, is governance. As AI increases what organisations can produce, it also increases the volume and speed of decisions that need to be made. More outputs, more scenarios, more situations where context matters. That creates a greater need for judgement, even if not labelled as such. In many organisations, the ownership of that judgement is still not clearly defined, which tends to create a gap between what is being produced and how it is being used. Those gaps tend not to remain abstract for long. They become visible when decisions have consequences.
None of this points to a fixed answer, which is probably worth keeping in mind. Coal held its position for a long time and then gradually lost it as the system evolved. The same pattern will play out here in different forms.
What seems to hold more consistently is the value of staying close to where decisions are made as things shift, rather than trying to anchor yourself to a particular category of work. That tends to be where context, judgement, and accountability come together, and where the shape of the work is harder to reduce to something purely mechanical.
