Insights / Topic
3 essays on this topic.
In March a company spent an average of $1.15 per million AI tokens. By September it paid 68 cents. Every budgeting process I know assumes input costs rise each year, and this input keeps falling in price.
Reviewing a piece of AI work costs $86, compared with $361 to do it yourself. It looks like a saving of three-quarters, and it is, but that $86 is priced in expert hours and has stayed the same even as the cost of generating the work has dropped a thousandfold.
A firm can now rent increments of thinking — bounded units of cognitive work, metered and priced like compute. That sounds like a pricing detail. It isn't. It changes what the cost structure of a firm is made of.