Korva is a scroll of the internet’s best ideas.
Don't mistake motion for progress
Metrics with holes
Every productivity metric has a hole. Lines of code looked like throughput, until one engineer's huge count came from porting a library. Significant lines of code came next, but updating a framework can produce less code with the same output. Time to land a PR was another candidate. Token maxing now looks like the old lines of code.
Outcomes over actions
Measuring tool usage measures the action. Starting from the problem to solve and a good way to measure that keeps the focus on the outcome, and a metric worth hill climbing still needs checking for whether it serves the outcome.
The listening tour
Beyond metrics, a listening tour works as teams adopt AI tooling. Senior engineers can say what is working and what is not, and they help multiply and scale it across the whole engineering team. Conversations like that spark ideas that dashboards do not.
Why do measures stop working once targeted?
Economist Charles Goodhart observed in 1975 that a measure tends to lose its reliability once it becomes a target. When people are rewarded for a number, they find the cheapest way to raise it, which may have little to do with the goal. Anthropologist Marilyn Strathern gave the idea its widely quoted form. It now goes by the name Goodhart's law.
Why was lines of code a poor measure?
Lines of code count volume, and good engineering often reduces it by deleting or simplifying. Andy Hertzfeld's account of the Apple Lisa project describes engineer Bill Atkinson filling in a form in 1982 with a figure of negative 2000 lines, after he rewrote a routine to be much shorter. Generated or ported code can add thousands of lines with little new function. The count rewards writing more instead of solving the problem.
How does hill climbing get stuck?
Hill climbing is an optimization method that keeps making whichever small change improves a score. It can end on a local peak, a spot better than its neighbors but lower than the best available. A metric can also be climbed past the point where it matches the real goal. Computer scientists add random restarts or other techniques to escape such peaks.
More from Lenny's Podcast: Product | Growth | Career
