When Intelligence Becomes an Input
AI turns intelligence from a scarce human capability into a purchasable input. The allocation question is what changes once that input becomes cheaper.
The important economic fact about AI is not that models can write, summarize, code, or reason. It is that a growing share of cognitive work can now be purchased in small increments.
That changes the allocation problem. A firm no longer chooses only between hiring, outsourcing, buying software, or leaving work undone. It can also buy machine intelligence, route tasks between different systems, and decide how much human review is economically justified.
The unit of analysis should therefore move from model capability to acceptable work. A model that is cheaper per token may be more expensive per accepted output if it fails more often. A frontier model may be worth its price if it reduces review time, lowers error risk, or makes a previously uneconomic task possible.
The allocation frame
The relevant question is not “which model is best?” It is:
What mix of capital, labor, software, and machine intelligence produces the desired output at the lowest total economic cost?
That total cost should include:
- inference and tool use;
- human setup and review time;
- failure and rework;
- latency and coordination;
- the cost of errors;
- the opportunity cost of scarce human attention.
This is why token prices are useful but insufficient. Tokens measure consumption. They do not measure production.
Why this matters now
Organizations are starting to make AI decisions with categories built for older technology. Some treat AI as software spend. Some treat it as labor substitution. Some treat it as a strategic transformation budget. Each framing captures part of the truth and misses the rest.
If intelligence becomes an input, management needs a way to compare that input against alternatives. The Allocation Review exists to develop that comparison in economic rather than theatrical terms.