When AI Masters Graded Problems, Problem Selection Becomes the Edge

AI models are getting extremely good at anything that can be turned into a clear loss function or well-defined, gradable task — schoolwork, LeetCode, standard engineering problems, and most routine coding.

In this environment, the highest-leverage human skills are shifting to the areas that are hardest to grade or automate within a single training loop:

•⁠ ⁠Problem selection — identifying which problems are actually worth solving

•⁠ ⁠Bringing high-level intuition and outside context to guide agents

•⁠ ⁠Allocating resources (tokens, time, compute) effectively

•⁠ ⁠Executing the "last mile" with taste, iteration, polish, and attention to detail

Traditional signals like LeetCode performance and standard system design interviews are becoming less predictive. What matters more is the ability to find signal in messy environments, choose ambitious problems, and turn raw model output into something extraordinary.

This Ledger Entry expands how readers think about career strategy and human value in the agentic AI era by showing that models are rapidly mastering anything that can be turned into a well-defined, gradable task — making skills like problem selection, resource allocation, bringing high-level intuition, and last-mile execution with taste the new sources of durable advantage.