Korva is a scroll of the internet’s best ideas.
$15,000 or $20 in monthly tokens
That spread shows up at Glean and at its customers alike. Glean never set a token budget. Like most companies, it let people figure out what they could do with the technology.
Everyone uses AI at least a little. The largest use of AI in the world is information seeking and question answering, so nearly every employee asks questions and gets some basic summarization and synthesis. Advanced use cases are limited to about 5 percent of the employee base.
Glean limited token-maxing dashboards that reward people for consuming more tokens, and judged them the wrong idea. As an AI-native company, its people are educated enough to use AI when they need it. Instead, every town hall has a section on the new AI agents teams are using to work differently.
Why does AI usage follow a power law?
A chat question uses a modest number of tokens, but an agent that plans, calls tools, and checks its own work can use many times more. Anthropic reported in 2025 that agents use about 4 times more tokens than chats, and that systems of several agents working together use about 15 times more. So a person who starts running agent workflows can leap far above someone who only asks questions. Usage in many fields follows the same shape, in which a small share of users accounts for most of the volume.
Why do usage leaderboards backfire?
Economists call it Goodhart's law: when a measure becomes a target, it stops being a good measure. Charles Goodhart stated a version of it in 1975 while writing about monetary policy. A dashboard that ranks people by tokens consumed measures spending, not value, so people can climb it without producing anything more. The same pattern has shown up with lines of code written and hours logged.
More from 20VC with Harry Stebbings
