Korva is a scroll of the internet’s best ideas.
AI school with a graduation
Memos written by AI
No meeting happens without a memo the night before, in the style Amazon uses, and the CEO spends about two hours each morning reading them. The goal is for 80% of reporting memos, like Monday's weekly growth report, to be written by AI, with AI surfacing the insights and a human section covering the topics that need discussion. That saves the team about four hours on that day.
The finance team's burn meeting memo, which covers cash and how the company is trending, follows the same target: 80% written by AI by the end of the month. The senior finance leader had been spending four to six hours a week building it.
Graduating from the school
A technical chief of staff goes to people and asks where they spend most of their time during the week, then the work gets automated for them. To graduate, an employee presents to the CEO what the method was before and what it is now, shows the output live, and reports how many hours it saves. A committee either graduates them or they stay in the school until they do.
The company has about 120 people and doesn't expect to cut any. Instead, some of its best people now run three AI agents each.
Why does Amazon run meetings on written memos?
Jeff Bezos banned slide presentations in senior meetings at Amazon in 2004. Attendees instead read a narrative memo in silence at the start of the meeting. Full sentences force the writer to show how the ideas connect, which bullet points allow a writer to skip. Amazon's memos are often up to six pages long.
How can AI draft a weekly report?
A script pulls the week's numbers from the company's database, and a language model turns them into a written summary that points out the biggest changes. A person reviews the draft and adds the topics that need discussion. Language models write fluently but can state numbers incorrectly, so teams usually check the figures against the source tables.
How does one person supervise three AI agents?
An AI agent is a program that takes several steps toward a goal on its own, such as searching, reading files, and writing results. The person sets the goal and reviews the finished work instead of doing each step. Because agents run in parallel, one person can have several tasks moving at once. The limit is how quickly the person can check what comes back.
More from Joys of Compounding
