AI-Generated Code Forces a Rethink of Software Quality and Developer Roles

As artificial intelligence takes over code generation, the nature of software itself is shifting. Phillip Mortimer, speaking at QCon London, argued that AI fundamentally changes how we should think about readability, testing, and the developer's role—pushing the industry toward write-only code that prioritizes functionality over human comprehension.

The Density Problem#

AI-generated code often becomes so tightly packed with logic that reading it for understanding becomes impractical. This echoes a principle from languages designed for extreme power—APL and regular expressions achieve remarkable performance through sheer syntactic density, sacrificing ease of reading. When code reaches this level of complexity, attempting to decipher it line by line becomes futile.

Rather than fighting this reality, Mortimer suggested embracing it. The solution lies not in making code more readable, but in shifting how teams validate and document it.

Tests as the Source of Truth#

In the write-only paradigm, comprehensive test suites become the actual specification. By examining input-output pairs, developers can understand what code does without reading the implementation itself. As Mortimer noted, this reverses traditional practices—tests stop being a safety net and become the primary documentation of system behavior.

This approach also simplifies maintenance. When AI-generated code proves difficult to modify or debug, rewriting it from scratch often takes less time than untangling the original logic. Teams keep the test suite and discard the implementation, using tests as the blueprint for a fresh implementation.

Scaling Code Review Beyond Humans#

Manual code review becomes a bottleneck when AI produces thousands of lines daily. Mortimer advocated replacing human reviewers with automated agents that can enforce standards, catch bugs, and raise pull requests continuously.

These agents can be:

• Embedded in CI pipelines to review each PR
• Configured with senior-developer heuristics and coding principles
• Run independently to identify repeated patterns, unused flags, and structural issues across the codebase
• Even tasked with self-reviewing AI output—a counterintuitive but effective adversarial approach

Additionally, self-healing systems that listen to observability platforms and automatically fix common issues represent the next frontier. An AI agent monitoring alerts can identify recurring problems and propose fixes without waiting for human intervention.