Expect current AI deployments to look primitive in retrospect
Current AI deployments mostly fit the technology into existing workflows; treat today's designs as transitional and expect later shapes to differ fundamentally.
The first motor vehicles were designed to look and work like horse-drawn carriages. Steering tillers rather than wheels, bench seating arranged for a driver holding reins, coachwork that matched the horse-drawn idiom. It took decades for the technology to evolve into something fundamentally different — and for the infrastructure around it (roads, fuelling, insurance, zoning, licensing) to evolve alongside. The horseless carriage was an intermediate form, shaped by what it was replacing rather than by what it could ultimately become.
Most current AI deployments are in the same phase. The technology is being fitted into existing workflows, business models, roles and org charts. Copilots sit inside the documents, meetings and tools people already use. The assumption is that AI adds capability to the existing shape of the work.
What the analogy implies
Two things follow from taking the analogy seriously.
First, today’s AI designs should be treated as transitional. The interfaces, the procurement patterns, the assumption that AI is a feature of existing software rather than a different kind of work arrangement — none of these are likely to be the shape that AI takes at steady state. Organisations that architect irreversibly around current designs are building on a form that will not hold.
Second, the genuinely disruptive AI patterns will not look like better versions of what exists. The motor vehicle’s transformative effect was not in being a faster horse; it was in enabling suburbs, freight networks and car-centric urban form. The AI analogue is not a better chatbot but a different arrangement of who does the work and how. AI as a labour service bypasses the adoption problem is an early example of that: the disruptive form is not a tool inside existing firms but an outside provider delivering outcomes the firm used to sell.
How to use the heuristic
When evaluating an AI deployment decision, it is useful to ask: which features of this design are specific to the present phase, and which would survive a genuine reorganisation of the work around AI’s capabilities? The first category is worth investing in lightly. The second is worth investing in deeply.
The heuristic is not an argument for paralysis. Horseless carriages were still useful. But they were not worth betting a firm on, and the firms that grew fastest were the ones that saw past the transitional form quickly.