A new benchmark asks whether coding agents can act like staff engineers. I like the question, because most tests ask about one task and a staff engineer's job is never one task.
It is the whole set of responsibilities: deciding what to build, how the pieces fit, what to leave alone. A benchmark built on real situations can show how close models are to carrying that.
The same list tells you where you stand. Whatever the score says, the responsibilities that remain are yours, and they are worth practicing on purpose.
The score belongs to the model; the responsibilities that remain belong to you.
Kay
P.S. Try the design-with-blocks plugin, which walks you through designing a real application from its building blocks. https://systemthinkinglab.ai/design-with-blocks