The 9-month vs 24-month ramp claim is interesting but I wonder how it holds up when the junior moves to a new codebase or domain. The traditional ramp isn't just "learning to code" - it's building mental models of how systems fail, developing judgment about when to ask questions, learning to read legacy code that makes no sense.
AI tools are fantastic at collapsing the "how do I syntax this" problems. But I've seen juniors get stuck when the AI confidently suggests something that compiles but misses the business context entirely. That kind of failure mode is harder to debug than a stack trace.
The economics argument makes sense though. If you're losing 36% before breakeven anyway, reducing that to 15% is huge even if the quality distribution stays the same.
The 9-month vs 24-month ramp claim is interesting but I wonder how it holds up when the junior moves to a new codebase or domain. The traditional ramp isn't just "learning to code" - it's building mental models of how systems fail, developing judgment about when to ask questions, learning to read legacy code that makes no sense.
AI tools are fantastic at collapsing the "how do I syntax this" problems. But I've seen juniors get stuck when the AI confidently suggests something that compiles but misses the business context entirely. That kind of failure mode is harder to debug than a stack trace.
The economics argument makes sense though. If you're losing 36% before breakeven anyway, reducing that to 15% is huge even if the quality distribution stays the same.