Every emerging tech has to yield to the Muck Layer
Most knowledge jobs - regardless how online they seem - have a moat that protects them from being replaced by pure automation, a deep “muck layer”. This muck layer is a sort of Polanyi Paradox - tasks we understand and navigate intuitively but have a hard time verbalizing - that machines are not suited for:
interpersonal and interorganizational politics,
gut feelings,
EQ,
unclear fail / success scenarios,
real-world complexity of systems, both digital and atom-based
organizational and network inertia
Because of the muck layer, I suspect that the large swath replacement of human employees with AI that’s been predicted over the last few years will not happen.
Here are the signals:
Forward deployed engineering heating up proves the muck layer - this happened with implementation consultants in the 2010s. Forward deployed engineers are a sign that the promise of AI automation is, to some extent, failing for the time being. AI is supposed to be empowering, and automation in a box, and yet all frontier labs are growing forward deployed engineers on staff to provide what the LLM can’t - reliable performance.
Complexity of harnesses to get useful and safe results; an average business owner / manager is not going to take this on
I (and many others) have written about Jevons Paradox and how it changed writing code. Human tasks change, but demand on the overall volume of human effort isn’t decreasing en masse.
Head on the line: an important roadblock in deploying AI is liability for when things go wrong, not just executing whatever the task is. If the automated response is a mistake, someone has to hold the bag. And as long as there’s a liable human in the loop, that human will need an ecosystem of humans around them.
Messy data and unclear procedures: sort of a Polanyi Paradox on an enterprise scale and not a new problem, something Digital Transformation efforts have been working on for at least a decade.
Real world friction, like orchestrating humans
Caveats
There’s a high number of researchers that will tell you this is only relevant insofar as humans are involved in the economy; it’s possible that economy and capital take off independently, leaving labor behind; “work” as in economically profitable activities, will be taken over completely by machines, which effectively eliminates the Muck Layer. This is a viable take if for no other reason than because a lot of smart people see it as viable. But after 200+ hours dedicated specifically to the question of “how likely is this?” I am less and less convinced that this is a real scenario. Elements of the economy, sure, like biotech R&D or maybe the boilerplate-code-writing industry. The entirety of the economy? No way.
Of course that doesn’t mean that the career landscape will stay static, jobs will change as they always have. One of the earliest signals is the growth in AI engineers, forward deployed engineer roles. But knowledge work as a profession is not going anywhere.
Related datapoints:
“LLMs are less like sophisticated, strategic advisors and more like enthusiastic newcomers who skimmed the textbooks, plugged into the ether of buzzwords, and enthusiastically offer their advice.” HBR