A recent analysis from Greptile argues that market forces will eventually prioritize maintainable AI-generated code despite current trends toward low-quality output. The report highlights token economics as a key driver for reducing software complexity and improving long-term reliability. This shift could redefine how development tools prioritize maintainability over raw output volume.
Key Details
The term slop describes unwanted, mindlessly generated content flooding the internet, a concept Simon Willison helped popularize. Recent data from Greptile indicates lines of code per developer grew significantly as AI tools became standard practice. Developers now ship more code, but this escalation raises alarms about software brittleness and maintenance costs.
According to the 2025 State of AI Coding report, average lines of code per developer rose from 4,450 to 7,839. Median pull request sizes increased 33% over a single year, reaching 76 lines changed. Individual file changes became 20% larger, suggesting agents are generating denser code blocks.
Analysis of vendor status pages shows outages have steadily increased since 2022, implying software is becoming more fragile. Andrej Karpathy noted that agents often bloat abstractions and copy paste code blocks without aesthetic consideration. Engineers report that while quantity is high, the quality of generated maintenance tasks remains a significant concern for teams.
What This Means
John Ousterhout argues in Philosophy of Software Design that complexity is the primary enemy of well-designed systems. Good code requires less context to understand, which translates directly to token efficiency for language models. Simple code reduces the compute and token costs required to generate and maintain software over time.
Greptile contends that markets will not reward slop in coding because complexity scales exponentially in cost. AI models that win will help developers ship reliable features fastest by adhering to simple, maintainable structures. Competition among model labs will eventually force a shift toward generating higher quality output to stay viable.
The industry is currently navigating a messy phase of innovation focused on getting AI to work rather than optimizing abilities. Once code generation becomes ubiquitous, economic incentives will start to take effect more strongly. Future competition will likely favor models that prioritize simplicity and maintainability over raw output volume.
