AI Productivity Gap: The New Performance Divide

Here is the reality of the divide:

1. The Velocity Delta: The devs leveraging a full AI workflow—Claude Code, Copilot, and custom prompt engineering—are clearing tickets at 2-3x the speed of the purists. They aren't necessarily "better" engineers in terms of architectural thinking, but they eliminate the friction of boilerplate and API discovery.
2. The "Quality" Argument: The holdouts claim AI introduces bugs or "lazy" code. While true if you're blindly hitting Tab, the heavy users are actually spending more time on edge-case testing because the initial build happened in seconds, not hours.
3. The Promotion Trap: This is where it gets messy. Management sees the output metrics (tickets closed, PRs merged) and rewards the velocity. The "craftsmen" who insist on writing every line from scratch are starting to look slow, not meticulous.

The real question is whether we are rewarding actual engineering skill or just the ability to steer an LLM agent. If a dev can ship five features in the time it takes another to ship one, the cost-to-value ratio makes the decision for the company.

Why are some senior devs still resisting this? Is it a genuine quality concern, or just a fear that their "years of experience" are being commoditized by a prompt?