The AI Dev Schism: The Psychological Loss of Making

The Distinction Between Making and Commissioning

Generating software via AI is fundamentally different from building it by hand because it replaces the act of creation with the act of commissioning. While AI tools can produce functional end products rapidly, the psychological fulfillment derived from "making"—the process of problem-solving, manual implementation, and deep engagement with the craft—is lost when a user simply prompts an LLM to generate the result.

For developers who value the process of construction, using an AI to write code feels less like authorship and more like management. As Brian Hall notes, the skill involved in prompting is the skill of effectively asking someone else to make something, which is a leadership or managerial function rather than a technical one. This creates a "schism" in the developer community between those who prioritize the end product (the "what") and those who prioritize the craft of implementation (the "how").

The "Gray Area": Compilers vs. LLMs

One common counterargument is that developers have always used abstractions, such as compilers, which translate high-level code (like C or Rust) into machine code that the human did not write. However, there is a critical distinction between a compiler and a Generative AI:

Deterministic Translation: A compiler performs a mathematically precise, deterministic translation of the developer's explicit instructions. The developer retains full agency over the logic; the compiler is merely a tool, similar to a hammer driving a nail.
Agentic Decision-Making: An LLM does not just translate; it "reasons" and makes autonomous design decisions. It may bake in behaviors the user did not explicitly request, shifting the user's role from a creator to an editor or a manager.

When a developer writes C code and uses a compiler, they feel they "made" the program because the logic originated entirely from their own mental effort. When they prompt an LLM, the AI makes the micro-decisions, leading to a sense of alienation from the final output.

Impact on Learning and Cognitive Connection

Manual coding fosters a deep, nuanced understanding of a system that AI-generated code often bypasses. This lack of "cognitive friction" can lead to several long-term effects:

Loss of Nuance: Building by hand requires the creator to grapple with every edge case and architectural decision, creating a mental map of the codebase. AI-generated code is often viewed as a "black box" that the user must reverse-engineer to understand.
Memory Decay: There is a reported difference in retention between hand-written and AI-generated code. Developers find it easier to recall the structure of a 100k+ line codebase written years ago by hand than to remember the details of a project recently "vibe-coded" with an AI.
Atrophy of Skill: Relying on AI for the "logical" parts of programming may lead to the atrophy of critical problem-solving regions of the brain, making the act of returning to manual coding feel "painful" or like "cardio."

Community Perspectives: Product vs. Process

Discussion among developers reveals a divide in how "making" is defined based on the user's goals:

The Product-Oriented View

Some argue that the end result is the only metric of value. From this perspective, AI is a force multiplier that allows individuals to build projects that would previously have required a full team. They view the prompt as the primary creative act and the AI as a highly efficient contractor.

The Process-Oriented View

Others maintain that the value of a side project is the experience of creation itself. For these developers, the efficiency of AI is actually a detriment because it removes the struggle that makes the eventual success rewarding.

The "Authorship" Debate

Drawing parallels to the art world, some compare AI coding to the workshops of Renaissance painters or the conceptual art of Marcel Duchamp, where the "idea" and the "curation" are the primary creative contributions. In this framework, the developer acts as a director or editor, and the resulting work is still "theirs" because it reflects their vision and taste.

Summary of the AI Dev Schism

Aspect	Manual Making	AI-Assisted "Vibe-Coding"
Primary Role	Creator / Craftsman	Manager / Editor
Core Skill	Implementation & Logic	Vision, Judgment & Prompting
Fulfillment	Derived from the process	Derived from the outcome
Understanding	High-fidelity, nuanced	Low-fidelity, surface-level
Speed	Slow, iterative	Rapid, immediate