The Case Against AI Coding: Craft, Cognition, and the Risk of Skill Collapse
The integration of AI into software development is often framed as an inevitable evolution—a move from manual labor to automated efficiency. However, a growing counter-current argues that this transition isn't just a change in tooling, but a fundamental threat to the nature of engineering. The debate centers on whether coding is a means to an end (the software product) or a craft where the process of creation is inseparable from the quality of the result.
This exploration examines the arguments against AI-assisted coding, ranging from the psychological impact on skill development to the systemic risks of economic and environmental instability.
The Psychology of Craft and Learning
For many engineers, coding is not a chore to be optimized, but a craft to be enjoyed. When the act of writing code is outsourced to a prompt, the intrinsic satisfaction of problem-solving is lost. This is not merely a matter of preference; it is a matter of cognitive development.
Recognition vs. Recollection
A critical distinction exists between recognizing a solution and being able to recall or derive it. AI provides immediate answers, which can create an illusion of competence. True skill development requires the "struggle"—processing information, making mistakes, and resolving contradictions within one's mental model.
As noted in the source material, mundane tasks are often the foundation of critical knowledge. When these are skipped, engineers lose sight of the constraints of their medium. This leads to a homogenization of output where everything looks the same because the creators no longer understand the boundaries they are supposed to be pushing.
The Feedback Loop of Skill Decay
There is a dangerous paradox in using AI: to effectively guide an AI, you must be more skilled than the tool. However, if you stop exercising those skills because the AI is "good enough," your ability to judge the quality of the output decays. This creates a downward spiral:
- Outsourcing: You use AI to handle complex tasks.
- Atrophy: Your manual problem-solving skills decline.
- Dependency: You become less capable of spotting "confident hallucinations."
- Acceleration: Frustration with your own decaying skills makes AI more tempting, further accelerating the decay.
Systemic Risks: From Skill Collapse to Model Collapse
Beyond the individual, there is a systemic risk to the profession of software engineering. If the industry stops hiring junior engineers because AI can handle entry-level tasks, the pipeline for senior engineers is severed.
The End of Capability
AI models are trained on human-generated data. If the economics of the industry shift such that fewer humans are writing original, high-quality code, the training data for future models will increasingly consist of AI-generated output. This leads to "model collapse," where hallucinations compound and the system degrades. More alarmingly, we may face a "total skill collapse" where the human expertise required to fix the models no longer exists.
Software as a Team Sport
Code is often viewed by management as an asset, but in practice, every line of code is a liability—a potential bug or source of confusion. The true asset is the shared understanding among a team of humans who wrote the code and understand the "why" behind every decision.
AI can produce code, but it cannot produce understanding. When a team relies on AI-generated blocks, they are essentially introducing legacy code into their system from day one—code that no one on the team truly owns or understands.
The Economic and Environmental Mirage
Critics of the AI boom point to a staggering disconnect between the cost of infrastructure and the actual profit generated. The current AI ecosystem is described as a "tower of debt," involving landlords, colocation companies, and AI labs, all betting on a revenue stream that has yet to produce significant profit.
The Resource Cost
The environmental toll is equally stark. The energy requirements for running human-brain-equivalent models are astronomical. Estimates suggest that replicating the synaptic capacity of a human brain would require billions of dollars in GPUs and gigawatts of power—resources that may be fundamentally unsustainable when compared to the cost of employing a human white-collar worker.
Counterpoints: Tooling vs. Craft
Not everyone views AI as a threat to the craft. Many developers argue that AI is simply the next step in a long line of abstractions, similar to the move from Assembly to high-level languages or from Vi to modern IDEs.
Common counter-arguments include:
- Ergonomics: AI removes the "boring and repetitive typing," allowing developers to focus on higher-level architecture.
- Augmentation: Using LLMs for security analysis or navigating complex, non-pythonic APIs (like PyQGIS) can act as a "smarter rubber duck" or an improved search engine.
- The "Tractor" Analogy: Just as the tractor didn't kill farming but changed its scale and nature, AI may simply change the scale of software production.
Conclusion
The tension between AI and manual coding is a conflict between two different philosophies: software as a product and software as a craft. For those who view it as a product, AI is an efficiency gain. For those who view it as a craft, AI is a shortcut that bypasses the very experiences that make an engineer competent. Whether the industry trends toward total automation or maintains a core of human expertise, the risk remains that in the pursuit of speed, we may be erasing the knowledge required to maintain the systems we build.