As I said, Alea AIcta est.

A growing concern is emerging regarding the extensive use of AI in software development. The argument goes that as AI generates more code, developers gradually distance themselves from the codebase they are supposed to understand and maintain. This issue is not without validity, as the developer often enters the process late, reviewing thousands of lines of code generated by AI.

This leads to a loss of the mental model of the system, making code review difficult, akin to searching for a needle in a haystack. While AI-assisted development presents challenges, the problem may not solely lie in the AI itself but in how it is utilized. AI should not be regarded merely as an API invoked for specific tasks. Instead, seen as a conversational interface, AI can serve as a valuable collaborator throughout the development process.

By engaging in continuous dialogue, developers can provide context, revisit decisions, explore alternatives, challenge suggestions, and gain explanations. This approach transforms AI into an integral part of the development workflow, akin to an interface through which developers think, explore, decide, and build. In AI-Native development, this distinction is crucial.

AI does not merely generate code; it can be a component through which developers construct the application. The difference between handing an AI agent a task and collaboratively developing an application step by step through continuous interaction with AI is significant. In the latter scenario, AI becomes part of the process, rather than an isolated generator whose output must be understood afterward.

Ownership remains essential, even when AI-generated code is involved. Through incremental development, developers continually read and comprehend the existing codebase, understanding the rationale behind each component, its role in the system, and its implications for future steps. Consequently, AI-generated code does not necessarily equate to "someone else's code."

As developers integrate and extend AI-generated code sequentially, their mental models of the codebase can expand alongside it. Ownership fades only when developers cease to comprehend the decisions they adopt, rather than when they stop personally turning them into syntax. AI may impede the development of certain skills, such as mastering a specific framework's syntax, but it can simultaneously expedite the acquisition of other competencies.

Developers utilizing AI might spend less time on syntax-level implementation and more time contemplating architecture, systems, products, users, technology layers, and decision consequences. This shift in capabilities does not inherently signify progress or stagnation. Instead, it prompts a crucial question: Are developers merely learning less due to AI's influence, or are they consciously deciding to focus on different aspects, while simultaneously acquiring new skills?

Lastly, it is vital to recognize that AI can create a distance between developers and the process of writing code. Reduced time spent on syntax-level implementation and manual function assembly may result in a diminished connection to the coding process. However, this distance from the code does not equate to a detachment from the system as a whole.

Despite spending less time directly at the coding level, developers can still maintain a strong understanding of the overall system, thanks to AI's collaborative nature.