Day 162 of the Daily · New Technology & Superintelligence

The Falling Price of a Draft

5 min read · 4 principles · apply it by lunch

Morning. The part of the work that got cheap was never the part that cost you.

The question

If producing a first draft in your business got ninety per cent cheaper tonight, which line of your accounts would actually move? For most operators none of them would, because nobody has yet split the cost into the part that produces and the part that checks.

The idea

The AI Index, published each year by the Stanford Institute for Human-Centered AI and edited by Nestor Maslej, exists to replace argument about this technology with measurement. Its 2025 edition put a number on what everyone had been sensing. Querying a model performing at GPT-3.5’s level on the MMLU benchmark cost twenty dollars per million tokens in November 2022; by October 2024 the same capability cost seven cents, a fall of more than two hundred and eighty times in under two years. Underneath that, hardware cost per unit of performance fell roughly thirty per cent a year while energy efficiency improved around forty. The cost of training a frontier model, meanwhile, went the other way, into the tens and hundreds of millions. Every figure here ages in months, so the year matters as much as the number. And what hasn’t fallen at all is the cost of checking the output, wiring it into your systems, and getting people to work differently. The trap is arithmetic: the falling line is the one with a published number against it, and the costs that stayed put are the ones nobody has ever put on a page.

In the Daily · 4 principles and a template

The Falling Price of a Draft is Day 162 of 365.

In the Daily, each of its four principles is explained in full, with a one-page template to apply it the same day. A new framework arrives every morning at 7am for a year, across all fifteen courses of the FutureMBA curriculum.

$29 a month · cancel any time · 7-day money-back guarantee