Codegen is not productivity
antifound.com
March 15, 2026
18 min read
🔥🔥🔥🔥🔥
47/100
Summary
Generative AI, including large language models (LLMs), can produce significant amounts of code. However, measuring productivity solely by lines of code generated is considered an inadequate metric for assessing software development output.
Key Takeaways
- Lines of code generated by LLMs do not fundamentally change the measurement of programmer productivity, which has historically been deemed a poor metric.
- Developers spend most of their time on activities other than coding, indicating that code generation is not the primary work of a programmer.
- Programming is primarily about managing complexity and representing abstract ideas, rather than just writing code for machines to execute.
- The generation of code is not the bottleneck in software development; it has never been the sole focus of the development process.
Community Sentiment
Positives
- GenAI can significantly speed up the coding process, allowing for rapid regeneration of large portions of code when requirements change, which can enhance adaptability in development.
- The use of architecture decision records can help mitigate overly optimistic outputs from GenAI, ensuring that generated code aligns with sound design principles.
Concerns
- The reliance on GenAI may shift the uncertainty in coding from the implementation phase to the specification phase, potentially leading to more expensive mistakes later in the development process.
- There is a concern that the speed of code generation is prioritized over quality, with teams treating AI outputs as final products rather than drafts that require thorough review.
