Korva is a scroll of the internet’s best ideas.
The weekly prompt
The prompt asks what feedback his manager, Ami Vora, would give based on everything he did or didn't do that week. Claude draws on her public writing about product, internal discussions, and what she has recently posted in Slack to work out her priorities.
Direct reports
He runs a version on his direct reports, too: Claude looks at what each of them did this week against team goals and OKRs, along with discussion transcripts, and suggests key takeaways and feedback to give.
How good it is
The quality is hit or miss. It works like soft coaching, with a coach who occasionally says something clearly off, and other times catches something that makes him glad he saw it.
Why does a named role shape Claude's answers?
Language models learn from large amounts of text, so they pick up the typical views and phrasing associated with a named role. A prompt that names a specific person or job shifts the tone and priorities of the answer toward that role. Studies of persona prompting have found that it changes style more reliably than it improves factual accuracy.
How did OKRs spread beyond Intel?
Andy Grove developed the objectives and key results method at Intel in the 1970s. Venture capitalist John Doerr, who had learned it there, introduced it to Google in 1999, and he later wrote about it in his 2018 book Measure What Matters. Objectives and key results give an agent a measurable yardstick to compare a week's activity against.
Why is AI feedback sometimes clearly off?
Slack captures only part of how a manager forms a view, since hallway talk, calls, and private chats never reach it. A model filling those gaps can produce confident feedback built on a wrong premise. Anthropic researchers also reported in a 2023 paper that language models trained with human feedback sometimes tailor answers to what they think the user wants to hear.
More from Lenny's Podcast: Product | Growth | Career
