Greenfield vs. Brownfield: Why AI Coding Demos Do Not Match Reality

When you hear that AI writes 90% of the code at Anthropic, it sounds like the game is over for programmers. But Google has AI generating over 30% of their code, and their actual engineering velocity only went up by about 10%. That gap tells you something important about what coding actually involves.

The Greenfield Illusion

In software, "greenfield" means building something new from scratch. Clean slate, no history, no baggage. Almost every impressive AI coding demo you have seen is greenfield. Start fresh, build a thing, ship it. AI looks unstoppable in that context.

Helen from Artificiality speaking to camera with bookshelves behind her, title overlay reading "Greenfield v brownfield and AI abundance"

"Brownfield" is the opposite. You are working inside a system that has been growing and accumulating decisions for years, sometimes decades. There are millions of lines of code written by people who left years ago, full of weird choices that turn out to be load-bearing. Nobody remembers why a function exists until you delete it and everything breaks.

Helen explaining the brownfield reality, subtitle reading "There's millions of lines of code"

The Real Bottleneck

Almost every actual company on earth operates in brownfield environments, including Google. The code was never the hard part. The hard part is knowing which code to write, where it fits, what will break, and what the system needs to survive contact with 20 years of accumulated decisions.

That is why Google's 30% AI-generated code translates to only 10% velocity improvement. The bottleneck is not typing. The bottleneck is understanding.

Helen making the key point about what code needs to survive, subtitle reading "need to survive, what it needs to do to survive"

Beyond Software

This distinction applies to every job. Every role has a greenfield version and a brownfield version. The greenfield version is clean, describable, and fits in a manual. AI handles that well. The brownfield version is tangled up with history, context, relationships, and institutional knowledge that lives in many different places.

That brownfield work is what actually happens in most organizations. And as AI handles more of the greenfield tasks, the brownfield capabilities become more valuable, not less.

Key Takeaways

Greenfield (clean slate) is where AI coding demos shine; brownfield (existing systems) is where real companies operate
Google's 30% AI-generated code only produced a 10% velocity improvement because the bottleneck is understanding, not typing
Every job has a greenfield and brownfield version, and AI handling the greenfield makes brownfield skills more important