Maintenance.

AI-generated code that quietly ignores a repo's established conventions — reinventing helpers, picking the wrong library, and using off-house naming and error-handling — so the codebase drifts toward generic, internet-average style.

AI assistants confidently emit APIs, libraries, and idioms that were current years ago but are now deprecated, removed, or superseded, because their parametric knowledge is frozen at a training cutoff and the old pattern is still the statistically most common one in their corpus.

AI assistants hedge against failures that can't happen, wrapping already-safe code in redundant null checks, dead guard clauses, and catch-all try/except blocks that add complexity without adding safety.

AI assistants bake environment-specific values — URLs, ports, paths, timeouts, keys, and magic numbers — directly into logic instead of reading them from config or environment, because a literal is the most probable next token and the model lacks awareness of your existing config layer.