The individual anti-patterns are manageable. The Phantom Intern produces code nobody understands. The Congealing Slop makes the codebase harder to navigate. The Stagnant Miasma kills momentum. Each has a counter, and each counter can work when a team applies it early.
The trouble starts when the monsters arrive together.
The Phantom Intern is the pattern of accepting AI-generated code because it looks finished. It compiles, follows local syntax, and fills the ticket. It may also be structurally hollow: abstractions no one asked for, duplicated logic, or a new path through the system that only its author understands. That code creates the conditions for Congealing Slop. The repository fills up faster than people can review it, and the cost of understanding each change rises.
Once the codebase is difficult to read, the Stagnant Miasma takes hold. Engineers spend more time tracing behavior and less time changing it. Small fixes feel risky. Refactoring gets postponed because the team has no spare capacity. Then someone reaches for AI to unblock the work, which produces more code that is easy to accept and hard to understand. The loop is already running.
The Miasma also gives the Scope Creep Kraken room to grow. When a team cannot ship features at a useful pace, the backlog becomes a place to put every plausible improvement. AI makes that list cheaper to produce. Suggestions arrive quickly, each one reasonable in isolation. Taken together, they form a workload no team can finish. New work displaces cleanup, and the accumulated cleanup makes new work slower.
This is why a code review rule aimed at one problem can matter to several. The patterns are connected by the team's shrinking ability to pay attention. The Phantom Intern's code is harder to review because the Slop has already eroded the team's understanding. The Slop is harder to remove because the Miasma has drained the time and confidence required for refactoring. The Miasma is harder to break because the Kraken keeps adding work that looks more urgent than maintenance.
The organizational patterns from Volume 2 add another layer. A team drowning in generated code may not have time to notice the Cost Singularity building in the background. The Cost Singularity stays untouched because the Compliance Black Hole has turned governance into a document exercise rather than a working habit. The Autonomy Mirage then makes it easy to skip the review that would have caught the Phantom Intern's contribution before it shipped.
That chain does not require anyone to be careless. Every step can look sensible at the time. A developer accepts a small generated change. A manager defers cleanup to meet a deadline. A governance group requests another document. A team adds a meeting or a tool to handle the resulting coordination problem. The system gets worse through ordinary decisions, not one dramatic failure.
The combination problem also changes how teams should diagnose trouble. If code quality drops, the answer may not be a better lint rule. If delivery slows, another productivity tool may not help. The visible symptom is often downstream from a different pattern. A backlog full of AI-generated suggestions may be a response to slow delivery, while slow delivery may come from code nobody feels safe changing.
Start by identifying which two or three monsters are active in your environment. Look for the reinforcement loop, then interrupt one connection. If the Phantom Intern is feeding the Slop, use the Explain It rule: before AI-generated code is approved, the developer must explain what it does and why it belongs in the system. The point is not to demand a lecture. The point is to catch code that nobody actually understands.
That one rule will not clean up the whole repository. It can slow the flow of hollow code long enough for review to become useful again. From there, the team can choose a small refactoring target, reduce the backlog of speculative work, or restore an ownership boundary that has gone blurry.
You do not need to slay every monster. You need to stop them from helping each other.