It's a variant of AI slop specific to agentic coding and automation tools, where a long unsupervised run compounds small mistakes into a messy end result. It's usually a sign the agent needed tighter scoping, more checkpoints, or human review partway through rather than a single end-to-end run.