Coding AI without deterministic outcomes (www.aha.io)

In a recent discussion, an engineer expressed frustrations about working with AI, particularly in the context of coding features that utilize large language models (LLMs). Unlike traditional programming, where bugs and outputs can be traced and rectified methodically, LLMs introduce a level of unpredictability that makes debugging and achieving desired outcomes challenging. This non-deterministic nature of LLMs complicates the sense of completion that programmers typically experience; the criteria for "done" becomes blurred, as outputs can always be improved upon, leading to an ongoing cycle of iteration without definitive closure. This evolution in coding practices signifies a shift in the software engineering landscape, wherein some engineers may thrive under the collaborative and ambiguous conditions presented by LLMs. The frustration with lack of clear determinism can lead some to seek roles that emphasize teamwork and creative problem-solving, such as engineering management or product management, aligning their work with personal strengths and satisfaction. As engineers adapt to these new workflows, the ability to effectively prompt and communicate with LLMs may emerge as a valuable skill, fostering a craft that may captivate those who enjoy interactive and human-like collaboration.