Code Isn't the Bottleneck Anymore

Two years ago, without AI or with primitive autocomplete tools, there was a natural limit to how much code a developer could produce. The speed of software was the speed of human typing, thinking, debugging. Code was the constraint.

That’s no longer true.

Where the Time Goes Now

Before AI assistants, I’d estimate 60-70% of my time went into writing code. Reading documentation, implementing features, fixing bugs, debugging. The actual work of translating ideas into software.

Today that number is closer to 30%. And it’s not because I’m working less.

The rest shifted to everything around the code:

  • Code review. More code produced means more code to review. And AI-generated code needs careful review because it can look correct while being subtly wrong.
  • Refining product definitions. When you can build fast, “what exactly should we build?” becomes the harder question. Vague specs that used to take weeks to implement now take hours, so the vagueness gets exposed immediately.
  • Quality and observability. Shipping faster means breaking faster if you’re not careful. More time goes into making sure things actually work in production, not just in the PR.
  • Communication. Aligning with product, design, stakeholders. The code is ready, but is everyone else?

The Bottleneck Moved

Here’s what I think is happening: AI solved the part everyone assumed was the problem.

For years, the industry optimized for developer productivity. Better IDEs, faster builds, more abstractions, hiring more engineers. All to produce more code.

Now code production is practically unlimited. And it turns out that was never the real constraint.

The real constraint is knowing what to build, making sure it’s right, and getting everyone aligned. Those processes didn’t get faster. If anything, they feel slower now because the contrast is so sharp.

What This Means

I don’t think this is a problem with a clean solution. It’s a shift in where the work lives.

Developers who only know how to write code will feel this pressure. The skill that mattered most, translating specs into working software, is now the easy part.

What matters more now:

  • Understanding the domain deeply enough to refine your own specs
  • Reviewing code critically, including AI-generated code
  • Communicating clearly with non-technical stakeholders
  • Knowing when to slow down and verify, not just ship

The code writes itself. The rest still needs humans.