The post explores why synthesis proves harder than analysis through the lens of calculus. Mathematicians have developed numerous calculi—from Church's lambda calculus to SQL's relational calculus—but traditional calculus refers to differential and integral variants. Differential calculus computes slopes at points via straightforward algorithms, enabling computers to calculate derivatives automatically. This process underlies LLM training through automatic differentiation techniques.

Integral calculus computes areas under curves, yet lacks universal algorithms. Instead, practitioners rely on specialized tricks for different function types, and some—like the Gaussian bell curve—yield no closed-form solutions. This disparity stems from a fundamental distinction: differentiation operates locally, requiring only neighborhood information at a point, while integration demands global knowledge across entire intervals. Local problems consistently outrank global ones in difficulty.

This insight translates to operational domains like SREs, where incident response frequently presents synthesis challenges. SREs must reconstruct how system components normally interact to diagnose current failures—a fundamentally global problem. Because synthesis exceeds analysis in complexity, engineers cannot achieve superhuman understanding of every component. Instead, cultivating expertise in system interactions becomes essential for resolving complex incidents effectively.