The Economics of Difficulty
A machine arrives on the floor and everyone gathers around it.
The workers want to know where they stand, and what they are supposed to do while the machine runs. The consultant wants to know how the money moves now — which step everyone has to route through, and where a toll can be charged.
Whether or not the machine is clever, it can cost jobs and rebundle skills. Systemic change does not wait for evidence at the level of the individual worker. Belief moves first, then action, and the value chain is re-cut on the strength of a forecast — well before anyone has established what the machine can actually do, and sometimes instead of ever establishing it.
It also happens when the machine does exactly what was promised.
Typing pools existed because a clean page was hard to produce — one error cost the whole sheet. Word processing made correction free, and the difficulty was repackaged: a fraction of it into the machine, the rest into everyone else's day, a few seconds at a time that nobody counted. Nothing outperformed the typist. The system was rearranged around her, and the new arrangement had no place where difficulty was worth buying on its own.
This is why reskilling misreads the threat. It treats a change in the system as a gap in the person. What the machine removes is not the skill but its scarcity, and scarcity belongs to the arrangement, not to you. Learn another hard thing and you have bought a few years. What survives repackaging is work whose worth was never difficulty.
Difficulty is a property of tasks, and tasks are what machines absorb. Absorb the task and the difficulty goes with it.
Being answerable is not a task. When the system changes and something goes wrong, someone has to be wrong, and someone has to be accountable. That is an exposure, not a difficulty, and there is nothing in an exposure to absorb. It can only be placed.
Sangeet Paul Choudary's read is that this is a business. Capability commoditises fast; accountability does not commoditise at all. So price it — the warranty, the indemnity, the SLA, the party contractually answerable when the output is bad. Value migrates to whoever absorbs the customer's risk, and that position holds long after the model underneath it is a commodity.
Cory Doctorow's read is that the same structure, built cheaply, produces a scapegoat. Put one reviewer in front of output generated faster than they can check it, at a pace they do not control, and the error has not been caught. It has been assigned. And the person it is assigned to is never the person who bought the machine.
The difference is whether the cost of being wrong lands on the vendor's revenue or on an individual's performance review.
References
Sangeet Paul Choudary, Reshuffle: Who Wins When AI Restacks the Knowledge Economy (FutureFirst Books, 2025).
Cory Doctorow, The Reverse Centaur's Guide to Life After AI: How to Think About Artificial Intelligence—Before It's Too Late (Farrar, Straus & Giroux / Verso, 2026).