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

A recent discussion in the software development community emphasizes the critical responsibility of programmers to deliver thoroughly tested code, especially in an era where AI tools like large language models (LLMs) are increasingly used. The conversation highlights a concerning trend of junior engineers relying on LLMs for generating untested code, which burdens their colleagues during the review process. This underscores the importance of manual and automated testing: developers must not only test their code manually but also integrate automated tests to ensure reliability, thereby establishing a strong foundation for their contributions. The article also notes the rapid advancement of coding agents that can execute and test code autonomously, such as Claude Code and Codex CLI, which marks a significant shift in how engineers can leverage AI in software development. However, it emphasizes that human accountability remains paramount; while AI can facilitate testing, the responsibility for ensuring that code is proven to work lies with developers. As LLM capabilities evolve, the challenge for engineers is not just in writing code, but in providing proof of its functionality, thus pushing the envelope for quality and accountability in software development.