¿Prefiere leer en español? Leer en español →
Traducción hecha con IA: algunos términos pueden no traducirse bien y puede haber errores.
The case against
Where we might be wrong.
An idea is only as good as the best argument against it. These are the strongest we have found. Disagree with us? Write a response and we’ll publish it next to ours.
Coordination costs
Splitting work into many small pieces adds hand-offs. Coordination costs may grow faster than the savings.
Coase (1937); Williamson, Nobel lecture (2009): the more specific work is to one firm, the more it tends to stay inside.
Some work resists description
Judgement, relationships and taste don’t fit neatly into input, transformation and output.
Our own open question.
Know-how nobody wrote down
Much of what makes work good is never recorded, so a capability card can miss it.
Our own open question.
Lost apprenticeships
If small tasks go to software, how do people learn the job? Junior roles are where expertise is usually built.
Our own open question; the conversation on education and the new skills will take it on.
Delivery is failing today
Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027.
Gartner press release, 25 June 2025: gartner.com.
AI isn’t robust yet
In Anthropic’s shop experiment, AI shopkeepers improved but stayed open to manipulation; in one benchmark the best agent completed 24% of office tasks when first published (December 2024) and 30% in a September 2025 revision.
Anthropic, Project Vend phase 2 (Dec 2025); Xu et al., TheAgentCompany (arXiv 2412.14161).
Gamed numbers
Once a measure becomes a target, people hit the number without delivering what it stood for.
Goodhart (1975, published 1976), on monetary statistics; the popular wording is commonly credited to Strathern (1997).
Just a new vocabulary?
Describing work as capabilities may add nothing that process maps and job design don’t already give us.
Our own research question: what would show it adds nothing.
A registry of vague labels
A weak classification could turn a precise-looking registry into a catalogue of ambiguous labels.
Our own research question.
People pay the price
Breaking jobs into pieces can weaken bargaining power and professional identity, and shift risk onto individuals.
Our own research question; employment law questions go to lawyers.
Autonomy is the industry’s goal
Most maturity models aim at “no human interference”. Our view, that a named person answers for the work, runs against that.
Digital Twin Consortium maturity model (Nov 2024).