Despite the recent focus on chat applications, code workflows might be a better beachhead application for generative AI. Why?
Code is (mostly) deterministic. Code generation deals with a more structured and deterministic environment than natural language chat applications, which are often ambiguous and context-dependent. AI-generated code can be tested for correctness (e.g., unit tests, static analysis), whereas chat applications often require human intervention to clarify misunderstandings or handle complex, nuanced conversations.
Economic value. Code is a non-rivalrous good — its use by one person does not diminish its value or availability to others. Add that with the zero marginal distribution cost and the fast-growing developer population, and you get extreme returns to developer productivity increases. Some chat workflows have this property as well, but not all do.
Direct impact on software development. Generative AI is coming to every text box, but integrating it into larger natural language workflows will be more challenging. On the other hand, generated code fits easily into any developer workflow (it’s just code). In addition, AI-generated code can be tested and validated by the same AI system, identifying errors and suggesting fixes before the code is integrated into the project.
Scalability. Code generation can enable the rapid development of large-scale software projects by automating repetitive tasks and streamlining workflows. In contrast, chat applications are limited by the need for human input and interaction, which cannot be easily scaled to handle massive amounts of data or complex tasks.