AI Makes Code Write-Only and Disposable, Requiring New Development Approaches

The Write-Only Nature of AI-Generated Code

At QCon London, Phillip Mortimer presented the idea that artificial intelligence is making code write-only. He defined write-only code as being too complex or ill-structured for anyone but its author to comprehend or modify. This concept applies to AI-generated code, which often exhibits extreme density, similar to languages like APL or regular expressions.

Tests as Documentation and Code Disposability

Mortimer suggested that for such dense code, tests become the primary documentation. Developers understand the code's function by examining input-output pairs from extensive test cases, rather than directly reading the code itself. Furthermore, AI-generated code is often disposable; its complexity makes debugging difficult, so rewriting the code from scratch, while retaining the tests, is frequently more efficient than attempting modifications.

Challenges in Code Review and Proposed Solutions

The sheer volume of AI-generated code makes human line-by-line review impractical, as humans quickly become a bottleneck. To address this, Mortimer proposed automating code reviews and developing detailed tests. He also suggested that software should be capable of self-healing, with AI coding agents monitoring observability platforms, aggregating alerts, and initiating pull requests to fix identified issues.

Decoupling Intent from Implementation

Mortimer highlighted the decoupling of intent from implementation in the age of AI. With AI using natural language, developers can become more portable across different programming languages. This shift allows developers to focus on creative aspects of software development, as AI handles the implementation details.