Has AI Ruined Software Development? No - It Shifted the Work to Specs
Has AI Ruined Software Development?
80% of the time is now writing specs and constraints to contain the AI, not coding. And honestly, that might be an improvement.
The Work Shifted, It Didn't Disappear
Before AI coding agents, a senior developer's day looked like: 30% writing code, 30% reviewing code, 20% debugging, 20% meetings. Now it looks like: 60% writing specs and constraints, 20% reviewing AI output, 10% fixing edge cases, 10% meetings.
The total effort hasn't changed much. But the nature of the work has fundamentally shifted. Instead of writing implementation code, you're writing precise descriptions of what the code should do, what it shouldn't do, and how to verify it did the right thing.
Why Specs Are Actually Harder
Writing a good CLAUDE.md file or a detailed spec for an AI agent is harder than most people expect. You have to think about:
- Edge cases upfront - the agent will find them if you don't
- Negative constraints - what the code must NOT do is as important as what it should do
- Verification criteria - how do you know the output is correct without reading every line?
This is the kind of thinking that used to happen unconsciously while coding. Now it has to be explicit. That's not easier - it's just different.
The Silver Lining
Here's what people miss about this shift: writing specs is closer to architecture and design than writing implementation code. The AI handles the boilerplate, the syntax, the plumbing. You focus on what matters - the constraints, the invariants, the business logic.
Developers who lean into this become better architects. The ones who fight it spend their days fixing AI-generated code that almost works - which is worse than writing it from scratch.
The career path is clear: learn to write specs, not just code. The developers who thrive in 2026 are the ones who accepted that the job changed.
Fazm is an open source macOS AI agent. Open source on GitHub.