The trend in 2026 will be for software developers using "AI" coding assistants to become more and more reliant on them as their own ability to comprehend and reason about programs atrophies.
This is make them less able to spot problems in the output, of which we *know* there'll still be many. That problem isn't going away.
The deterioration of software reliability will accelerate as models are trained on more and more of the "slop".
It's a doom loop.







