The article emphasizes the responsibility of software developers to deliver code that they have verified to work, particularly in the era of AI-assisted coding. It critiques the trend of junior engineers relying on large language model (LLM) tools to produce untested pull requests (PRs), which places an undue burden on their colleagues during the review process. Developers should not only write code but also provide proof of its functionality, ensuring that it meets established standards before submission. Proving code works involves two essential steps: manual testing and automated testing. Manual testing requires developers to actively verify that their changes produce the desired results, often through direct interaction with the system. The author suggests documenting this process with terminal commands and outputs to enhance transparency during code reviews. Automated testing, which has become more accessible with LLM tools, should accompany any code changes. This step adds a layer of assurance, as the test should fail if the code doesn’t perform as expected. The rise of coding agents like Claude Code and Codex CLI has changed how developers interact with code. These tools can execute code and help ensure its correctness. However, developers still need to guide these agents to perform manual tests and create automated tests. The author argues that human oversight is crucial, as computers cannot be held accountable for their outputs. Ultimately, the value lies in contributing code that is demonstrably functional, not just in generating large patches.