Korva is a scroll of the internet’s best ideas.
Agents are the primary audience
Earnings transcripts
More people probably consume earnings transcripts through agents than listen to the audio or read the transcript itself. These high-value pieces of the web have flipped to agent first, even though the communication around them hasn't. A public company CEO giving an earnings report today has reason to make sure that what gets said will be transcribed and interpreted correctly by agents, and not only by the people listening.
Developer docs
Developer documentation has gone the same way. Customers building AI solutions use AI to do it, so their agents read the docs and the code in the SDKs, and humans rarely fumble around docs pages. For that reason an agent is the primary audience of those docs, and they are tested with an agent.
The picture is a parallel web for agents that sits next to the human one, and publishers write for both.
Why do earnings calls draw so much transcription?
In 2000 the SEC adopted Regulation FD, which requires public companies to share material information with everyone at the same time. Companies responded by hosting open earnings calls that anyone can join. Services transcribe these calls within hours, and the text feeds directly into financial databases. A call produces a public record that many systems read.
How do developers make docs readable by agents?
In September 2024, Jeremy Howard proposed a file called llms.txt, placed at the root of a website. It lists the most important pages in a plain Markdown format that a model can read with little effort. Several documentation platforms and developer tool companies have adopted it. The idea resembles robots.txt, which has told crawlers where to go since 1994.
More from Training Data
