# Code is cheap, quality is not AI coding agents have made generating code nearly free. Multiple agents can work in parallel while a single human oversees them. This disrupts decades of engineering intuitions built around code production being expensive and time-consuming. The critical distinction: **generating code is cheap; producing good code remains expensive.** Good code requires correctness, proper testing, error handling beyond the happy path, simplicity, maintainability, documentation, security, accessibility, and observability. Agents can draft all of these, but humans must still ensure the standards are met. This inversion has practical consequences. Old instincts like "that's not worth the development time" should be reconsidered. If the downside of trying something is just token consumption and the upside is discovering a valuable implementation, the rational move is to experiment. [[Simon Willison]] recommends reversing the old scarcity mindset: try things asynchronously with agent prompts, evaluate the results, iterate. The flip side is equally important. Since coding is now cheap, the old justifications for accepting [[Technical debt]] weaken. Tasks that were "conceptually straightforward but too time-consuming" (API redesigns, nomenclature cleanup, duplicate consolidation, large refactors) can now be handled by agents in the background. There's less reason to let quality slide. The risk is treating cheap code as an excuse to skip quality. That's the [[Unreviewed AI code anti-pattern]]: generating volume without verification. Cheap code only produces value when it passes through human judgment. ## References - https://simonwillison.net/guides/agentic-engineering-patterns/code-is-cheap/ ## Related - [[Agentic Engineering]] - [[Technical debt]] - [[Unreviewed AI code anti-pattern]] - [[Cognitive debt]] - [[AI Agents]] - [[Simon Willison]]