Why Software Engineers Are Divided on AI - The 5x Gain Is Not Where You Think
Why Software Engineers Are Divided on AI - The 5x Gain Is Not Where You Think
Half of software engineers say AI tools make them dramatically more productive. The other half says AI generates garbage code that takes longer to review than writing from scratch. Both are right - they are just measuring different things.
The Writing Trap
Engineers who focus on AI for code generation often end up disappointed. The AI writes plausible code, but it misses edge cases, introduces subtle bugs, and does not understand the broader system architecture. Reviewing AI-generated code can take as long as writing it yourself, especially for complex logic.
This is where the skeptics have a point. If you measure AI productivity by "lines of code generated per hour," you are measuring the wrong thing.
Where the 5x Actually Comes From
The engineers who report massive productivity gains are using AI differently. Their top use cases:
Code review acceleration. Reading a 500-line diff and understanding what changed, what might break, and what patterns are violated. AI does this faster and more consistently than humans, catching issues that reviewers miss when they are fatigued.
Codebase navigation. Finding where something is implemented, understanding call chains, tracing data flow. Instead of grep-and-read loops, you ask a question and get an answer with file references.
Documentation and explanation. Understanding legacy code, writing docs for existing systems, explaining complex logic to new team members.
Boilerplate and glue code. Not the interesting algorithmic code - the config files, API wrappers, test scaffolding, and plumbing that everybody agrees is boring.
Why the Divide Exists
Engineers who use AI for creative problem-solving are underwhelmed. Engineers who use AI for the tedious parts of the job are thrilled. The tool is the same. The application is different.
The productivity gain is not in replacing the thinking. It is in accelerating everything around the thinking - the reading, navigating, reviewing, and scaffolding that surrounds the actual design decisions.
Fazm is an open source macOS AI agent. Open source on GitHub.