When I first designed the framework for The AI Developer's Field Guide, I borrowed the alignment system from tabletop RPGs. You know the one: Lawful/Neutral/Chaotic on one axis, Good/Neutral/Evil on the other. I adapted it to software: Structured/Pragmatic/Chaotic × Verified/Speculative/Hallucinated. Nine cells, two axes, a tidy grid.
It was a mistake.
The Problem
Readers told us the alignment grid was confusing. Not because the concepts were wrong — "how far from reality is this person operating?" is a real question — but because the grid itself was a barrier. You had to learn it, remember it, and then decode labels like "Pragmatic Hallucinated" or "Ornately Plausible" every time they appeared. Some character sheets used the standard grid labels. Others invented custom ones that mapped to the grid "intentionally." That sentence in the book — "Later character sheets occasionally riff on these nine slots with more colorful alignment labels" — was an apology disguised as a feature description.
The grid was trying to do something simple — describe how a person relates to reality under pressure — and making it complicated. So we killed it.
What Replaced It
Each archetype now has an Operating Mode: a single plain-English sentence that tells you what you need to know. No grid, no axes, no decoding.
| Archetype | Operating Mode |
|---|---|
| Bulldozer of AI Confidence | Charges ahead without questioning AI output |
| Shield Bearer of Stagnation | Plants feet and refuses to move |
| Daredevil of AI Disaster | Chases every new tool without settling |
| Archmage of Analysis Paralysis | Evaluates every option exhaustively before committing to any of them |
| Illusionist of AI Complexity | Reaches for the most sophisticated solution available, whether or not the problem needs it |
| Visionary of Grandiose Plans | Operates at the roadmap level, where every present task is a placeholder for a future that never arrives |
| Shadow Patcher of Ephemeral Solutions | Fixes the immediate problem now, leaves coherence for later |
| Bandit of Borrowed Brilliance | Assembles impressive-looking solutions from other people's work and presents them as original insight |
| Outlaw of AI Plunder | Treats licenses, terms, and governance as friction to be bypassed rather than constraints to be respected |
| Faithful Automator | Trusts process and AI output with equal faith |
| Ritualist of Redundancy | Follows every process, including the ones they invented |
| Doomsday Archivist | Preserves everything, discards nothing, ships rarely |
You can read any of those in a second and know what you're dealing with. That's the whole point.
The Character Sheets Got Thinner Too
The same feedback told us the stat tables were trying too hard. Every archetype had 15+ rows, and many of them were forced jokes — "Primary Damage Type: Confident implementation" or "Typical Pull Request Size: Large enough to feel heroic and just small enough to avoid immediate intervention." Funny once, maybe. Funny twelve times in a row, across four class chapters? Not really.
We cut the joke rows: Passive Trait, Party Role, Primary Damage Type, and Typical Pull Request Size are gone. What remains is the information that actually helps you recognize and work with the pattern:
- Class and Anti-Class (what archetype this is)
- Operating Mode (the one-liner)
- The seven core abilities (Discernment through Ship Velocity, with a note on how this archetype scores)
- Signature Anti-Pattern (the mistake they repeat)
- Weakness (what exposes them)
- Countermeasures (what helps)
That's a profile you can actually use in a code review or a retro. The rest was decoration.
What Didn't Change
The seven core abilities are still there. The classes (Fighter, Wizard, Rogue, Cleric) are still there. The anti-patterns, the artifacts, the monsters — all unchanged. The Working in Teams sections we added to each anti-pattern are staying. The framework is the same. We just stopped trying to make it look like a Dungeon Master's Guide and started making it read like a field guide.