As I have been saying (over and over), the employment apocalypse rests on a misunderstanding of how enterprise systems actually evolve. It assumes that the existing software order can be overturned by enthusiasm and marginal cost curves. It assumes that the operating expenses of an AI agent are transparently lower than those of a salaried employee, and that organisations will therefore reorganise overnight. This confuses technical capability with institutional transformation, and it's embedded in the AI business plans and Opex. The apparent naivety of these AI gurus outside their core expertise is stunning. Even Elon recently commented that the current crop of AI guru predictions shows they've never built anything; the very reason he wants to put data centres in space is that he can't get the electricity to build them on Earth. This mirrors my reasoning. Modern enterprises are not loose collections of tasks waiting to be automated. They are bureaucratic structures built on procedures, accountability chains, compliance regimes, and risk management. Software is embedded inside that structure. It encodes reporting lines, audit trails, permissions, regulatory obligations, and financial controls. Replacing a human is not the same as replacing a line item. It is replacing a node in a web of responsibility. There is an assumption of instant convertibility: that firms will discard their SaaS infrastructure and replace it with agents because token costs appear attractive. This ignores the deeper cost, supervision, validation, integration, security hardening, model monitoring, and failure management. The visible API price is only the surface. Beneath it sits an apparatus of control. Rationalisation won't move at the speed of demos. It moves at the speed of institutional legitimacy. Procurement cycles, internal politics, retraining, regulatory clearance, and risk committees govern adoption. A system that handles payroll, compliance, or client records cannot be replaced because a podcast claims efficiency. The biggest question on AI take-up is not whether AI can perform a task. It is whether AI can be incorporated into an existing bureaucratic order with equal reliability, calculability, and accountability at a lower total cost. Only when it satisfies those criteria does it become rational for the organisation to restructure itself around it. There will be a change (there always is, I have been looking at software migration for 40 years ). There will be displacement (again, that has been consistent for 40 years). But the fantasy of instantaneous labour replacement assumes a world without institutional inertia. Modern capitalism is built on administration. Administration does not dissolve because a new tool appears. It absorbs, evaluates, and reorganises only then.