The ancient Greeks had a word for something that can heal in one hand and poison in the other: 'pharmakon.' More than two millennia before Claude, Perplexity, or Pangram existed, Socrates used the idea to describe writing itself.
Plato recounts an Egyptian myth in which the god Theuth presents writing to King Thamus as a remedy for memory and wisdom. Thamus is not convinced. Writing, he answers, will produce forgetfulness because people will rely on external marks instead of their own memory.
That same tension now sits at the center of generative AI. Used by someone with knowledge and judgment, it can help retrieve, organize, test, and revise ideas. Used as a substitute for those capacities, it can produce fluent prose without understanding.
The comparison is not just philosophical. It is operational. The story frames AI writing as a practical tradeoff between speed, cost, and intellectual ownership, and the market has already voted with its feet.
The Associated Press made that trade in 2014, automating corporate earnings stories and expanding coverage from roughly 300 companies per quarter to 4,400. Bloomberg did something similar around the same time, publishing headlines with the help of machine learning starting in 2015 and using automated technology to assist with roughly a third of its content as far back as 2019, according to The New York Times.
That is the real pressure point for any institution that produces content or ideas. The choice is no longer whether AI can be kept out. It is how visible the intervention should be, and what is being surrendered in exchange for the efficiency.
The piece also points to a more immediate market signal. When Stanley Druckenmiller published a Wall Street Journal op-ed on the bond market in August, readers running it through Pangram got a 100% AI-generated verdict. Druckenmiller confirmed it within hours. The piece carried no disclosure, and Paul Gigot defended the silence rather than require one.
That episode captures the hard edge of the new normal. Capital rewards clear rules, and institutions that blur the line between original thought and automated assistance invite distrust. If AI is helping produce the work, the question is not only whether it is useful. It is whether the reader, the client, or the taxpayer is being shown the truth about how the work was made.
The old Greek warning still applies: every useful tool can also displace the faculty it was meant to support. In a competitive market, that bargain may be unavoidable. But it is never free, and the cost is not just measured in time saved. It is measured in judgment, credibility, and the long-term value of the work itself.
