Why Economics Will Kill AI Slop and Save Good Code

While AI tools currently encourage a high volume of messy code, economic incentives will eventually shift the focus back to quality. Good code is cheaper to maintain and requires fewer compute resources, making it the more sustainable choice for businesses. As the AI market matures, the models that produce simple, reliable code will win by allowing developers to ship faster at a lower cost.

Key Points

• AI coding tools have significantly increased code volume and PR density, leading to more brittle software and frequent outages.
• Good code is defined by simplicity and shallow interfaces, making it easier to understand and modify than complex AI-generated 'slop.'
• Economic incentives drive the shift toward quality because maintainable code requires less compute and fewer tokens over its lifecycle.
• The current phase of messy AI generation is a temporary innovation hurdle that will be corrected by market competition.

Sentiment

The community is notably polarized. There is a slight lean toward agreeing that code quality matters, particularly among experienced engineers who have lived through technical debt crises. However, a substantial and vocal contingent argues that market realities consistently contradict the code-quality-as-virtue narrative, pointing to profitable companies with terrible codebases and the practical value of AI-generated 'good enough' solutions. The discussion is more nuanced than a simple agree/disagree split, with many commenters seeking context-dependent middle ground.

In Agreement

• Internal code quality (modularity, good abstractions) directly produces external quality (fewer bugs, better performance, evolvability), making it an economic advantage rather than mere aesthetics
• At scale, poor code creates concrete costs through infrastructure bills, outages, security vulnerabilities, and engineering velocity collapse
• Temporary code inevitably becomes permanent, so investing in quality from the start pays dividends when you cannot predict which code will survive
• The most productive engineering teams, like Jane Street, invest heavily in rewrites and quality yet still outpace less rigorous teams
• Feeding crappy code to LLMs produces worse outputs, creating a garbage-in-garbage-out cycle that compounds technical debt
• Good code is easy to replace while bad code is hard to replace, meaning low-quality code becomes disproportionately long-lived

Opposed

• Massive profitable companies like Salesforce, SAP, and Facebook succeed despite notoriously poor codebases, proving the market does not reliably reward code quality
• Enterprise software purchasing decisions are made by people who never use the software, structurally severing the feedback loop between quality and market success
• LLM-generated mediocre code democratizes software creation, enabling non-programmers to build useful tools that were previously impossible, and users of those tools do not care about internals
• Treating code as craft is a minority pursuit analogous to organic farming, while the real world runs on mediocre-but-functional solutions
• Humans have always written sloppy code and accumulated technical debt, so AI slop is not a fundamentally new problem but an acceleration of existing patterns
• Many vibe-coded internal applications are already saving significant time for real users, proving that mediocre code can deliver genuine value