# Coding Was Never a Bottleneck
Two product builders with experience serving both software engineers and consumer markets challenge the prevailing narrative around AI's impact on development velocity. They argue that coding speed has never been the constraint limiting software teams.
The piece questions the widespread assumption that AI coding assistants represent a genuine productivity breakthrough. While tools like GitHub Copilot and similar systems generate code faster, the authors suggest this misses the actual bottlenecks in software development. Engineering teams face real constraints elsewhere: architectural decisions, system design, testing, debugging, integration with existing systems, and the human coordination required to ship features at scale.
Historical context supports their skepticism. Compilers, version control, and frameworks already automated away much tedious coding work decades ago. Yet teams haven't seen proportional productivity gains. The missing piece isn't code generation speed. It's the thinking and planning work that precedes writing code.
The authors acknowledge their bias: they want AI productivity claims to be true. They build tools for engineers and scale consumer products, so faster development serves their interests directly. Yet their experience shows teams rarely ship features faster simply because they write code faster.
This reframes the AI productivity conversation entirely. Rather than celebrating the ability to generate 10x more lines of code, the real question becomes whether AI helps engineers make better architectural choices, catch bugs earlier, or reduce the coordination overhead of shipping software. Current tools excel at code completion and generation. They struggle with the deeper design and planning work that actually constrains shipping speed.
The argument doesn't dismiss AI's value in development. Instead, it redirects focus from raw coding throughput to where bottlenecks actually exist. Teams that treat AI as a code-generation speedup without addressing these deeper constraints likely won't see the transformative productivity gains they expect.