Member-only story
The Missing SDLC Layer in AI-Driven Development
Why AI coding assistants increased velocity — but quietly broke software engineering discipline
AI dramatically increased developer productivity.
But it also removed a guarantee software engineering relied on for decades:
Decision provenance.
We can still see the code. We can no longer reliably explain why that code exists.
The industry adopted AI faster than it learned how to engineer with it.
The Problem Nobody Talks About
We celebrate faster feature delivery thanks to ChatGPT, Copilot, and Claude. But under the surface, a critical gap is emerging.
Consider a real scenario:
A team generated a database migration using an LLM. A rollback happened a week later.
They ran the same prompt again.
The generated migration was different.
Subtle schema mismatch → data corruption → hours of incident response.
No one could answer:
- Which reasoning produced the deployed version?
- What context influenced the model?
