You know what's happening. Sarah merges her own pull requests. Marcus ships code nobody can explain. The team's "AI-powered code review" is an AI approving AI-generated code while a human watches the green checkmark and calls it due diligence.
You know it. You've thought it in standup. You've felt it during a demo. You've maybe even said something — carefully, indirectly, the way you say things when you don't want it to become a conversation about a person instead of a pattern.
Here's the problem: "Sarah merges her own PRs" is an accusation. "We have a Bulldozer problem" is a diagnosis. Both sentences describe the same thing. One starts a fight. The other starts a conversation.
That gap — between what you see and what you can safely say — is where every team using AI tools lives right now. The patterns are obvious. The vocabulary isn't.
The Pattern Is Not the Person
I wrote The AI Developer's Field Guide because I kept seeing the same anti-patterns across different teams, different companies, different tech stacks — and I couldn't talk about them without making it personal.
The RPG framing wasn't a gimmick. It was a solution to a specific problem: how do you name a behavior without naming the person doing it? "The Bulldozer of AI Confidence" is a pattern, not a performance review. Your best engineer can exhibit Bulldozer tendencies on a Tuesday and catch themselves by Wednesday. The class describes the behavior. The behavior is not the person.
Once you have the name, the conversation changes. "I think we're feeding the Scope Creep Kraken here" compresses a five-minute debate into a sentence. Everyone knows what that means. Nobody's being called out. The pattern is the subject of the sentence, not the colleague.
The Cards Work Because the Conversation Works
The book is the reference. It's the dictionary. It gives the patterns their full treatment — the character sheets, the stats, the counter-moves, the real-world costs. But a dictionary is not a conversation. You don't bring a dictionary to a retrospective.
The cards are different. You spread them on the table. Each person picks the three that match what they're seeing. Suddenly the team is talking about the Scope Creep Kraken and the Congealing Slop and the Phantom Intern — not about who's responsible for them. The cards make the patterns discussable. That's the activity. That's the whole thing.
I've watched it happen. Engineers who would never say "I think Marcus is shipping code he doesn't understand" will happily pick up the Phantom Intern card and say "this one's us." The card does the work. The RPG framing makes it safe to be honest about hard things.
What This Thing Actually Is
Slop Codex is a shared language for diagnosing what happens when AI tools meet human habits. That's it. The book establishes the vocabulary. The cards make it social. The bestiary on the website makes it browsable. The Phantom Intern chat makes it interactive. They all serve the same purpose: helping teams name what's going wrong so they can talk about it without it becoming a blame game.
I started writing it because it made me laugh. The patterns were painful, but naming them was funny in the way that good satire is funny — you read a chapter, wince, and say "yeah, that's us." The humor isn't decoration. It's the mechanism. The RPG framing works because it's the right amount of ridiculous to make a serious conversation possible.
The secondary reason I wrote it: I kept watching the same dysfunctions repeat across teams, and I wanted a way to capture them before they calcified into culture. "We have a Stagnant Miasma problem" is a thing you can act on. "Our architecture hasn't changed in three years and the AI keeps reinforcing the same patterns" is a thing that just sounds like a complaint.
The Hard Part: Saying It Without the Pain
Here's what I keep coming back to. Every team I've worked with has someone who ships too fast, someone who plans too much, someone who skips the guardrails, and someone who trusts the process more than the output. Before AI, these tendencies were constrained by human capacity. You could only ship so much code in a day. You could only plan so much before someone forced you to build.
AI removed those constraints. The person who ships too fast can now ship ten times faster. The person who plans too much can now generate ten times more analysis. The person who skips guardrails now has tools that happily comply. The person who trusts the process now has dashboards that confirm everything is fine.
The dysfunctions aren't new. The amplification is new. And the amplification makes the conversations harder, not easier, because the stakes are higher. When Sarah merges her own PRs, the damage is local. When Sarah merges her own AI-generated PRs at ten times the volume, the damage is structural. But saying "Sarah, stop" feels like an attack. Saying "we have a Bulldozer problem, and it's getting worse" feels like a diagnosis.
That's the thing I built. Not a book, not a card game, not a website — a way to say the thing you already know without it being completely painful. The patterns give you the words. The framing gives you the safety. The rest is just the conversation you were already trying to have.