If Writing is Thinking

People in large organizations rarely read long internal documents. The author of this post admits he sent year-long memos to thousands, counted on a handful of senior leaders to actually read them, and even created slide decks to force consumption. Yet months later he was still fielding questions because most of the team never absorbed the content.

He points out that the only thing people reliably read are short org memos, and even those need visual aids like org charts to prevent skimming. When the memo lacked a simple diagram, readers would wait for a separate graphic email before they would engage. The pattern repeats across domains: scientific papers suffer a reproducibility crisis because reviewers skim, and Wall Street analysts produce 30-page reports that few actually audit.

Enter generative AI. When AI writes a memo that the author never fully understands, the result is a compiled document no one reads end-to-end. Summaries generated on demand are often lossy or even hallucinated, yet busy executives still treat them as truth. This creates a feedback loop where decisions are made on shallow, potentially inaccurate information.

The practical takeaway for technical leaders is to stop assuming that a lengthy written artifact will be digested. Favor concise, visual communication, verify that key points are understood, and treat AI-generated text as a draft that needs human vetting before it guides strategy. Otherwise you risk building decisions on a house of cards.