Korva is a scroll of the internet’s best ideas.
Rank engineers by output, not tokens
Output over tokens
GPUs are expensive, but if engineers ship three times more, the math is clearly worth it. The point is to do it the right way: tie spending to specific outcomes, such as which tickets got done and which projects and initiatives moved.
A project comparison
Picture a project scoped for 18 months, headed to an outsourced contractor at a cost of 15 million dollars. Done internally with the team's own engineers and agents, it could cost 1 million dollars and finish in 3 months.
Speed of adoption
That is the kind of ROI worth defining clearly up front, before token budgets run away. Cognition works the same way with customers, since enterprise adoption typically takes 12 to 18 months and it aims to get deployed in about three.
Why does counting tokens mislead as a metric?
Economist Charles Goodhart observed in 1975 that a measure stops being reliable once it becomes a target. An engineer rewarded for tokens used can raise the number without producing anything of value. Output measures such as tickets closed and projects moved are harder to game. Even those need checking for quality.
Why do enterprise software rollouts take so long?
A large company typically puts new software through security review, legal review, procurement, and integration with its internal systems. Each step involves separate teams with separate schedules. Regulated industries such as banking add compliance checks on top. Adoption then has to spread team by team, and that is slower than the purchase itself.
More from David Senra
