Why Cheap Tokens Don’t Mean Cheap Value

A recurring theme running through this whitepaper’s Part II and Part III is that AI working capacity — tokens, agent-hours, orchestrated fleets of agents — has been getting dramatically cheaper to produce. Part IV takes the next, harder step: working capacity is not value. As the paper states plainly, value can be positive, zero, or negative, and no amount of cheap compute changes that [INTERPRETATION, drawing on Part IV’s framework].

One of the human analogies Part IV uses to make this concrete: the brilliant professor versus the street-smart entrepreneur. Both can have exceptional raw capability — one in analytical depth, one in practical judgment under uncertainty. But swap either into the other’s actual task and the output can be worthless or worse: the professor’s rigor can stall a fast-moving deal that needed a quick, good-enough call; the entrepreneur’s instinct-driven shortcuts can be flatly wrong on a problem that actually required careful analysis. Raw capability and fit-to-task are not the same thing — and that gap doesn’t close just because the capability got cheaper to generate [INTERPRETATION, Part IV, Section 25].

The same logic applies once AI agents, not people, are doing the work. Part IV argues that orchestration — running many AI agents in parallel — multiplies capacity, not value: bad judgment or bad information embedded in an agent’s instructions scales at machine speed across a whole agent fleet exactly as easily as good judgment does. A hyperscale-tier facility that drives cost-per-token toward zero says nothing about whether the resulting output is worth anything. As the paper’s own workbook note puts it, “a cheap, fast, wrong answer scaled across a multi-agent team is still cheap and wrong at scale” [INTERPRETATION, Release Asset #10, cited in Part IV].

The practical upshot, stated in Part IV rather than invented here: abundant, cheap AI working capacity increases — not decreases — the importance of human leadership, direction, and judgment about what that capacity should be pointed at. Cost curves are the easy arithmetic; deciding what’s worth doing is not, and this paper does not claim otherwise.

This is educational research synthesis, not investment, procurement, or policy advice.

Licensed under CC BY 4.0. Suggested attribution: Research synthesis and modelling by Valto Loikkanen, assisted by AI.