AI is turning software into 'write-only' disposable code, QCon speaker warns

Software engineering is facing a fundamental shift as AI-generated code becomes increasingly incomprehensible to the humans who deploy it. During a talk titled 'Complexity and Creativity in Software Engineering' at QCon London, Phillip Mortimer argued that the industry is moving toward a 'write-only' paradigm where implementation is a disposable commodity rather than a durable asset.

Mortimer describes 'write-only' code as logic so dense, arcane, or complex that it cannot be effectively read or modified by human developers. In this environment, the traditional process of human-led code review becomes a bottleneck, as AI can generate and rewrite code far faster than a person can debug it. Consequently, Mortimer suggests that the code itself should be treated as disposable, with the primary focus shifting away from the readability of the implementation.

The New Documentation

In this emerging model, the source of truth for how a system behaves shifts from the code to the test suite. Because the AI-generated implementation is often unreadable, extensive test cases must define the intended behavior. These tests effectively become the primary documentation for the software, ensuring that the disposable code meets the required specifications without requiring a human to parse the underlying logic.

A Shift in Engineering Value

This transition represents a pivot in the core value proposition of the software engineer. If AI has effectively solved the implementation problem, the engineer's role is no longer about the technical act of writing code, but about creative problem-solving and the rigorous definition of tests. By decoupling the intent—expressed in natural language—from the implementation, the human becomes the architect of the 'what' and 'why,' while the AI handles the 'how.'

The Future of Maintenance

As the industry adopts this disposable approach, the focus of maintenance will likely move toward automated verification and high-level orchestration. The goal is to move toward a world where the primary skill is not the ability to maintain a codebase, but the ability to define the constraints and outcomes that the AI must satisfy. What remains to be seen is how this shift will impact long-term system stability and the ability of organizations to audit their own software for security and compliance when the code is no longer human-readable.