The Gorman Paradox: Why AI Coding Assistants Fail in Production

Despite the hype surrounding AI coding assistants and promises of unprecedented productivity gains, a curious paradox has emerged: we see little evidence of AI-generated or AI-assisted software making it into production at scale. This phenomenon, which I'll call the "Gorman Paradox," suggests a hidden filter that prevents AI-assisted projects from reaching the advanced stage of real-world deployment.

Understanding the Constraints

Software development can be viewed as an iterative, goal-seeking algorithm with several key factors influencing its time complexity:

• Batch size: The amount of changes made in each iteration
• Feedback "tightness": How much uncertainty is reduced with each iteration
• Cost of change: Our ability to act on the received feedback

When analyzing these factors at scale, coding itself becomes a negligible component. The real heavy lifting is done by batch size, feedback loops, and cost of change.

The Firehose Problem

Attaching a code-generating firehose to development has already proven counterproductive when teams loosen their grip on batch sizes to gain initial benefits. Conversely, keeping AI on a tight leash means coding remains a minor factor even in high-performing teams, resulting in modest improvements at best.

This might explain why the productivity gains of AI coding assistants seem to vanish at scale. When user experience, reliability, security, and maintainability matter—attributes that are non-negotiable in production software—we're forced to consume AI-generated code in small, controlled increments.

The Hidden Filter

The constraint isn't the AI technology itself, but the development processes that govern how we integrate it. Teams showing modest gains using AI may have actually tightened their feedback loops to adapt to the firehose—a process that could be viewed as a stress test for development methodologies rather than evidence of AI's value.

This suggests that for real productivity gains to materialize, AI coding assistants are constrained to projects where production-quality attributes matter less: personal projects, prototypes, internal tools, and one-off applications. While undeniably useful for these purposes, such projects rarely make it out of the proverbial shed.

The Path Forward

As organizations continue experimenting with AI in development, the Gorman Paradox serves as an important reminder that technology alone cannot transform complex processes. The real value may lie in how we adapt our development methodologies to integrate AI effectively—focusing on the fundamental principles of batch size, feedback loops, and cost of change rather than relying on the technology itself to solve our productivity challenges.

The future of AI in software development may not be about replacing programmers but about creating systems that enhance human capabilities while respecting the constraints that have evolved over decades of software engineering practice.

Originally published on Codemanship