I have many, many thoughts about agentic coding. One of my main ones is that yes, you could get some great short-term gains with it. But it really feels that, like with “context rot”, the issues will eventually compound to a point where code is unworkable.
I guess, at that point, you have an agent rewrite from scratch against the same tests?
But will anyone know anything by that point?
Seems massively risky.