Rolling for Initiative: How AI Helped Me Learn to Run D&D

I’ve played D&D before — enough to know the basics, understand what a nat 20 means, laugh at the memes. But I’d always been a player. The Dungeon Master is the person behind the screen with the notes and the voices and the elaborate trap that nobody falls into. I was not that person.

Then I walked past a shelf at Target and spotted Dragons of Stormwreck Isle — a D&D 5e starter set, $24.99, includes a pre-made adventure and everything you need to start. I texted my group. They said yes. I bought it.

That was the right decision made for the wrong reason.

The Real Question

I assumed the hard part was learning the rules. It is not. The rules are a system, and systems can be looked up, referenced, and gradually internalized. A night or two with the booklet and you have enough to run a session.

The hard part is the world.

Dragons of Stormwreck Isle gives you an island, a threat, and a set of locations to visit. It does not give you texture. It doesn’t tell you the full history of why a particular NPC is who they are, or what the villain actually believes about himself, or what a halfling thief’s childhood in Luskan smells like at street level. Those details are the difference between a session your players enjoy and a session they remember.

I had no idea how to generate that texture at scale, on demand, before and between sessions.

Enter the Collaborator

I’d used Claude for work before — code, writing, the usual. But turning it into a D&D partner took about five minutes to feel completely natural.

The first thing I needed was a rules coach. Not a rulebook. I needed something I could ask “what actually happens if the rogue tries to steal from a myconid during a rapport spore session” and get an immediate, contextual, confident answer. No cross-referencing. No uncertainty about which page has the edge case. Just: here’s the rule, here’s how it probably plays, here’s what a reasonable DM would call.

That part was easy and immediately valuable.

The harder and more interesting part was building the world around my players’ characters. Before session zero, each player pitched their concept:

• A druid whose cartographer parents were killed in a gnoll raid, carrying his father’s compass
• A time-looping astral elf wizard who found a cursed book in a church restricted section
• A halfling rogue with ties to the Gilded Gallows thieves guild
• A fighter forged in the Crown Wars, built on loyalty and duty

Those were their seeds. My job was to make the world know about them before they knew about themselves.

Claude and I spent hours on this. Not generating generic backstory filler — actually working through the logic of each character’s history, figuring out what they’d suppressed, what they’d misunderstood about their own story, what a non-player character who’d crossed paths with them might know. That process changed the campaign. The kobold politics on Stormwreck Isle gained depth because I understood what the party would care about. The NPCs at Dragon’s Rest became more than quest-givers because I’d worked out who they were before I ever needed to write them a line of dialogue.

On the Recaps

I want to be honest about something: I’m not a great writer. I know when a story moment matters. I know when the table goes quiet because something landed. I don’t always know how to translate that feeling into prose.

After each session, I’d write up my notes — what happened, what surprised me, which moments had real weight — and Claude would take those notes and tell the story back. The session chronicles I shared with my players in our Discord timeline channel came from that collaboration.

AI shaped those recaps. Genuinely. The way Hobs’ looping time-vision is written — the fractured, recursive prose, the paragraph that repeats to make you feel what he feels — that’s Claude executing an idea I had about what that scene should do to the reader. The weight of Harlen’s one-shot kill that ended an ambush before it started: I told Claude what happened, Claude found the narrative frame for it.

That’s collaborative. Not a crutch. I brought the judgment — which moments mattered, what the characters actually said and did, where to slow down and let something breathe. Claude brought the execution.

And honestly? AI was capable of so much more than I used it for. Early sessions I was conservative, leaning on it for rules questions and light NPC polish. By Session 2, I was using it to develop full character dream sequences, faction motivations, inter-session plot threads. The quality of the story scaled with how much I trusted the collaboration.

What AI Can’t Do

It can’t see the look on a player’s face when an arrow lands and they realize the fight is already over. It can’t hear the table go quiet during a dream sequence. It can’t improv a response when your players decide to befriend the kobolds instead of fight them, or field a question about Runara’s past that you never prepared an answer for.

Those moments are entirely yours. And they’re the reason you do it.

Good improvisation at a D&D table is reading people, reading the energy in the room, knowing when to press and when to let a scene breathe. AI can prep you for that, but it can’t do it. The prep is what lets you focus on the live performance.

Where We Are Now

We’ve finished Dragons of Stormwreck Isle as written. The island is mostly saved, the party is level-appropriate for whatever comes next, and I still have session recaps to backfill before the full chronicle is written up.

But here’s what I know: the compass — Trenzor’s father’s compass, the one that’s been pointing the way since Session 2 — is going to be the artifact that drives whatever comes next. Not just in the fiction. Physically.

There’s something being built in my basement that Trenzor’s player knows nothing about. When it finally shows up at the table, the needle is going to spin, and everything we’ve built together — AI and DM and dice and friends — is going to land in one glowing, stupid-satisfying moment.

That’s the thing about learning something new. You don’t always know what you’re building until it’s in front of you.