129K Commits Later - Vibe Coding Is Just Coding
129K Commits Later - Vibe Coding Is Just Coding
When "vibe coding" first became a term, it carried a dismissive tone. You were not really coding - you were just prompting an AI and accepting whatever it generated. But after 129,000 commits where agents wrote the majority of the code, the line between vibe coding and regular software engineering has completely dissolved.
The Transition Nobody Noticed
It happened gradually. First, AI wrote boilerplate. Then it handled entire features. Then it started fixing its own bugs based on test failures. At some point, the human role shifted from "person who writes code" to "person who reviews code, sets direction, and decides what to build."
That is not a lesser role. That is a senior engineering role. The best engineers have always spent more time reviewing, architecting, and deciding than typing.
What 129K Commits Teaches You
At scale, patterns emerge that are invisible in small projects:
- AI-written code is more consistent - it follows the patterns you set without drifting over time
- Review quality matters more than writing speed - a bad review of AI code ships bugs faster than a slow human developer
- Tests become the specification - when agents write code, your test suite is the actual source of truth for what the system should do
- Style guides enforce themselves - linters and formatters matter more when 80% of code is machine-generated
The Review Bottleneck
The bottleneck has shifted from writing code to reviewing it. A single developer can direct five agents producing code, but reviewing that output at the quality level production requires is the hard part. This is why AI code review tools are the next critical piece of the stack - not to replace human review, but to filter the obvious issues before a human looks.
Stop Calling It Vibe Coding
If you are reviewing diffs, running tests, catching edge cases, and making architectural decisions, you are coding. The fact that an agent typed the characters does not change the intellectual work required. The tool changed. The craft did not.
Fazm is an open source macOS AI agent. Open source on GitHub.