Someone running a small business replaced a virtual assistant with a set of AI agents. The assistant charged $25 an hour. The agents cost $1,000 a month. At 160 working hours a month, that is a shift from $4,000 to $1,000. The same work. A quarter of the cost. He made a passing observation while describing it: wages tend to rise. Technology costs per unit tend to fall. He did not develop the point. He did not need to.

The two curves

Before the beliefs, the structure. Two cost curves govern this. One describes what it costs to employ a person to do a given category of work. The other describes what it costs to run an AI system doing comparable work. They move in opposite directions. Their crossing is the disruption.

The diagram is simple because the argument is simple. What is powerful about it is that neither curve requires dramatic change. Wages do not need to surge. AI does not need to become dramatically smarter. The curves need only continue in the direction they have been moving, in most cases for decades.

The question is: what is the minimum you need to believe for this to produce significant disruption at scale?

Five beliefs

These are not predictions. Each is a minimal proposition. Together they are sufficient for disruption at scale. Reject any one and the argument weakens. Accept all five and the disruption follows by arithmetic.

01 Belief
Already true
Wages trend upward over time.

This is one of the most durable observations in economic history. Labor organizes. Inflation corrects. Minimum wages move with political pressure. In both high-income and low-income markets, the cost of employing a person to do a given category of work rises over any sufficiently long period. You do not need to believe wages will rise fast. You only need to believe they will not fall permanently.

02 Belief
Already true
The cost to run AI work trends downward.

The cost per token of running a large language model has fallen by orders of magnitude in three years. Compute infrastructure is deflationary by structural necessity: chip density, energy efficiency, competition between providers, and the pure economics of scale all push the same direction. This curve does not require new breakthroughs to continue. It follows from the current trajectory of an already-established industry dynamic.

03 Belief
The operative condition
Substitution requires adequacy, not superiority.

The AI does not need to do the work better than the person. It needs to do it well enough that the substitution is rational at the price. A virtual assistant handling scheduling, research, and correspondence does not need to be replaced by something smarter. It needs to be replaced by something adequate at a quarter of the cost. The threshold for substitution is not excellence. It is the point at which the ratio of output to cost favors the machine.

04 Belief
The dynamic element
The adequate category expands as costs fall.

Today a task costing $25 per hour has crossed the threshold. As AI costs fall further, the crossover applies to work that currently costs $40, $60, $80 per hour. The capability of the AI may improve. But even if it does not improve at all, the pure price movement widens the category of work where substitution is rational. The disruption does not need AI to get smarter. It only needs AI to get cheaper. And cheaper is structurally guaranteed by Belief Two.

05 Belief
The adoption mechanism
The decision is made by one person with a spreadsheet.

Disruption at scale does not require cultural transformation, consensus, or a generational shift in attitudes. It requires a person with budget authority to look at two numbers and decide. That person is not the worker whose role is at risk. They are the owner, the manager, the founder. The decision is private, reversible, and immediate. There is no committee. There is no announcement. The role simply stops being filled. This is the fastest adoption mechanism in the history of technology deployment, and it is already in use.

The disruption does not need AI to get smarter. It needs AI to get cheaper. And cheaper is structurally guaranteed.

What follows

The compounding effect

Accept all five beliefs and the arithmetic is straightforward. Wages rise. AI costs fall. Adequacy is the bar, not excellence. The adequate category widens as costs fall. Decisions are made by individuals, fast and quietly.

This means disruption does not arrive as a wave. It arrives as a large number of individual, uncoordinated decisions, each rational in isolation, together constituting a structural shift in how work is organised. There is no moment where it begins. It is already in progress. The only question is the rate at which the two curves diverge further, and there is no structural reason to expect that rate to slow.

The jobs most exposed are not necessarily the lowest-skilled. They are the jobs that sit at the crossover: tasks that are routine enough to be replicated by an agent, but sufficiently paid that the cost gap is large. Virtual assistants, customer support tiers, junior research roles, data entry and processing, first-pass content drafting. Not because these people are replaceable in some human sense. Because the cost ratio has moved past the threshold where substitution becomes the default rational choice.

The categories above the current crossover are not safe indefinitely. They are safe until the price falls to meet them. Given Belief Two, that is a question of when, not whether.

Evolving Software Framework

The Evolving Software framework describes Feedback-Guided Direction as the mechanism by which systems move toward a measurable metric through iteration. That is what each individual substitution decision is: a feedback signal. One person, one spreadsheet, one rational choice. Multiplied across millions of organisations, those individual feedback signals aggregate into a system-level reorientation of how knowledge work gets done. No single actor is steering it. The direction emerges from the structure.

The thought exercise resolves here.

You do not need to believe AI will solve everything. You do not need to believe in AGI, or that the models will keep improving at their current rate, or that agents will handle complexity far beyond what they can today.

You need to believe that wages will not permanently fall. You need to believe that compute costs will not permanently rise. You need to believe that organisations respond rationally to price signals.

These are not bold beliefs. They are close to the most conservative positions available in economics. And they are sufficient.

The disruption does not require a revolution. It requires the continuation of two trends that have not reversed in living memory. That is the minimum you need to believe.