Good architecture compounds AI effectiveness — brute-force token consumption is the wrong optimization target¶
Insight: Experienced practitioners are shifting from examining individual lines of code to thinking strategically about architecture and system design when working with AI agents. Well-designed libraries with intuitive interfaces guide AI toward optimal solutions naturally, creating compounding benefits across the stack — whereas agents left to brute-force problems "grind problems into dust" by rewriting code repeatedly regardless of token consumption.
Detail: Webb describes spending less time reading code line-by-line and more time on library design and system architecture — reframing this shift as the productive form of "vibing." The key principle: when AI tools operate within well-designed systems, improvements propagate across the entire stack. This contrasts with the common pattern of throwing compute at poorly structured codebases. Willison's amplification of this perspective reinforces the emerging practitioner consensus that architecture is the primary leverage point for AI-assisted development.
Sources
- Matt Webb via Simon Willison — "Quoting Matt Webb" (2026-03-28)
Related: architectural-awareness-prerequisite-vibe-coding in vibe-coding.md — COMPLEMENTS