Your job is to deliver code you have proven to work (simonwillison.net)

Article summary

The article emphasizes the importance of delivering code that has been proven to work, rather than just submitting untested code for review. It highlights the need for both manual and automated testing to ensure that code works as intended. The author argues that relying solely on automated tests or LLM-generated code is not enough and that manual testing is essential to catch obvious bugs and usability issues. The article also discusses the role of coding agents and the need to learn how to get them to prove their changes work.

Main themes

• Code testing
• LLM-generated code
• Software development best practices
• Manual testing
• Automated testing
• Coding agents

What commenters say

• Manual testing is a crucial step in ensuring that code works as intended and should not be skipped, even if automated tests are in place.
• The use of LLM-generated code does not replace the need for human testing and verification, as it can introduce errors and misunderstandings.
• Some argue that automated testing can catch most bugs, but others counter that it is not a substitute for manual testing and human judgment.
• The importance of testing and verification is not limited to code quality, but also affects the usability and overall value of the software.
• There is a concern that the increasing use of LLM-generated code may lead to a decrease in code quality and an increase in bugs and errors.
• Others argue that LLM-generated code can be useful and efficient, but it requires careful review and testing to ensure it works correctly.
• The role of coding agents in software development is still evolving, and it is essential to learn how to effectively use them to improve code quality and productivity.
• Some developers believe that attaching screenshots or videos of working code to PRs can help prove that the code has been tested and works as intended.