AI tools and the need for concrete context

Sean Goedecke contrasts generic design advice (principles and patterns not grounded in a specific codebase) with concrete design (decisions shaped by the real code, constraints, and existing “prior art”), arguing you can’t meaningfully design software you don’t work on because you lack the context to make implementable calls. Generic advice has its place (greenfield work, company-wide guardrails), but in large, messy systems consistency matters more than isolated “good design,” because teams survive by reusing known patterns and keeping the codebase coherent. He’s skeptical of architect handoffs where designs ignore practical timelines and incentives reward complexity, and he notes AI coding tools behave like smart outsiders—useful, but prone to reinventing what already exists unless humans with deep context guide them.

Links:

Sean Goedecke’s article: “You can’t design software you don’t work on”

SOLID principles (overview)

Single-responsibility principle (SRP)

GitHub Copilot (product page)

GitHub Copilot code review (docs)

Claude Code (Anthropic product page)

GitHub adding Claude + Codex agents (The Verge)

Dead Code Podcast Links:

Mastodon

X

Jared’s Links:

Mastodon

X

twitch.tv/jardonamron

Jared’s Newsletter & Website

Episode Transcript

Hosted on Acast. See acast.com/privacy for more information.