The Bottleneck Is Never the Code

In The Goal, Eliyahu Goldratt introduces a simple idea that most people in software still haven't internalized: the throughput of any system is determined by its bottleneck. Not by the fastest part. Not by the part you just optimized. The bottleneck.

AI can write code 5X faster. Maybe more. Let's take that at face value and assume the code is good. Even assume it's 90% there.

Now what?

The Downstream Problem

Someone has to review that code. Someone has to test it. Someone has to merge it, resolve the conflicts, communicate the changes to the team, update the documentation, collect feedback from users, and fix the bugs that show up after shipping.

None of those steps got 5X faster. Most of them didn't get faster at all.

So now you have 5X the code flowing into a pipeline that was already at capacity. The reviews pile up. The test suite takes longer. Merge conflicts multiply. Bugs stack up faster than anyone can triage them. The team that was already stretched is now drowning.

The Real Bottleneck

The bottleneck of every software company is the same: the people. The team that builds it. The users who use it. The engineers who maintain it going forward.

Writing code was never the bottleneck. Understanding what to build was. Communicating why it matters was. Getting feedback, iterating, and shipping something that actually works in production was.

Speeding up the non-bottleneck doesn't help. Goldratt proved this decades ago. If your constraint is the oven and you buy a faster mixer, you just end up with more dough sitting on the counter.

More Code Is Not More Progress

It's tempting to equate output with progress. Five pull requests instead of one feels like five times the work getting done. But if each one creates review burden, introduces risk, and generates downstream tasks, you haven't made progress. You've created overhead.

The right question isn't "how much code can we produce?" It's "how much can our system absorb?"

If the answer is less than what you're generating, you don't have a speed advantage. You have a pile-up.

Use AI to write code faster, sure. But if you're not also investing in making the rest of the pipeline faster, all you're doing is moving the pain somewhere else. And that somewhere else is your people.

This post is brought to you by myself and Claude Opus 4.6.