Member-only story

The Real Bottleneck Is Not the Code

--

For a long time, we believed the hard part of software was writing it. That belief made sense when code was expensive: when every line required effort, recall, and careful construction. Today, that constraint is disappearing. With modern AI-assisted development tools, the act of producing code has become almost effortless. Entire functions, modules, even architectural structure can be generated in seconds. What used to be a bottleneck is now available at scale. Code is no longer the limiting factor in software development.

And yet, something no longer adds up. Systems are not becoming proportionally more reliable. Bugs have not vanished, vulnerabilities have not decreased, and complexity has not become easier to manage. If anything, the gap between what we can produce and what we can confidently understand is widening. This tension exposes a mistake that has been present in software engineering for decades, long before AI-assisted development: we misidentified where the real difficulty lies. The challenge was never primarily in writing code. It was in knowing, with precision, what the code is supposed to do.

It was always the specification.

Once you see this, much of the industry’s behavior starts to look less like engineering and more like adaptation. Modern development workflows are built around a cycle that feels natural but is fundamentally reactive: write code, test it, find defects, fix them, and repeat. Over time, we have refined this loop with better tools: unit testing frameworks, fuzzing…